Upload 47 files

.gitattributes

@@ -33,3 +33,7 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+demo_assets/depths/pumpkin.png filter=lfs diff=lfs merge=lfs -text
+demo_assets/input_imgs/pumpkin.png filter=lfs diff=lfs merge=lfs -text
+demo_assets/material_exemplars/cup_glaze.png filter=lfs diff=lfs merge=lfs -text
+fig/gradio_demo.png filter=lfs diff=lfs merge=lfs -text

DPT/EVALUATION.md

@@ -0,0 +1,60 @@
+### Genral-purpose models
+The general-purpose models are affine-invariant and as such need a pre-alignment step before an error can be computed.
+
+Sample code for NYUv2 can be found here:
+https://gist.github.com/ranftlr/a1c7a24ebb24ce0e2f2ace5bce917022
+
+Sample code for KITTI can be found here:
+https://gist.github.com/ranftlr/45f4c7ddeb1bbb88d606bc600cab6c8d
+
+
+### KITTI
+* Remove images from `/input/` and `/output_monodepth/` folders
+* Download `kitti_eval_dataset.zip` https://drive.google.com/file/d/1GbfMGuwg2VS06Vl75-_tB5FDj9EOrjl0/view?usp=sharing and unzip it in the `/input/` folder (or follow this repository https://github.com/cogaplex-bts/bts to get RGB and Depth images from list [eigen_test_files_with_gt.txt](https://github.com/cogaplex-bts/bts/blob/master/train_test_inputs/eigen_test_files_with_gt.txt) )
+* Download [dpt_hybrid_kitti-cb926ef4.pt](https://github.com/intel-isl/DPT/releases/download/1_0/dpt_hybrid_kitti-cb926ef4.pt) model and place it in the `/weights/` folder
+* Download [eval_with_pngs.py](https://raw.githubusercontent.com/cogaplex-bts/bts/5a55542ebbe849eb85b5ce9592365225b93d8b28/utils/eval_with_pngs.py) in the root folder
+* `python run_monodepth.py --model_type dpt_hybrid_kitti --kitti_crop --absolute_depth`
+* `python ./eval_with_pngs.py --pred_path ./output_monodepth/ --gt_path ./input/gt/ --dataset kitti --min_depth_eval 1e-3 --max_depth_eval 80 --garg_crop --do_kb_crop`
+
+Result:
+```
+Evaluating 697 files
+GT files reading done
+45 GT files missing
+Computing errors
+     d1,      d2,      d3,  AbsRel,   SqRel,    RMSE, RMSElog,   SILog,   log10
+  0.959,   0.995,   0.999,   0.062,   0.222,   2.575,   0.092,   8.282,   0.027
+Done.
+```
+
+----
+
+### NYUv2
+* Remove images from `/input/` and `/output_monodepth/` folders
+* Download `nyu_eval_dataset.zip` https://drive.google.com/file/d/1b37uu-bqTZcSwokGkHIOEXuuBdfo80HI/view?usp=sharing and unzip it in the `/input/` folder (or follow this repository https://github.com/cogaplex-bts/bts to get RGB and Depth images from list [nyudepthv2_test_files_with_gt.txt](https://github.com/cogaplex-bts/bts/blob/master/train_test_inputs/nyudepthv2_test_files_with_gt.txt) )
+* Download [dpt_hybrid_nyu-2ce69ec7.pt](https://github.com/intel-isl/DPT/releases/download/1_0/dpt_hybrid_nyu-2ce69ec7.pt) model (**or a new model** that is fine-tuned with slightly different hyperparameters [dpt_hybrid_nyu_new-217f207d.pt](https://drive.google.com/file/d/1Nxv2OiqhAMosBL2a3pflamTW39dMjaSp/view?usp=sharing)  ) and place it in the `/weights/` folder
+* Download [eval_with_pngs.py](https://raw.githubusercontent.com/cogaplex-bts/bts/5a55542ebbe849eb85b5ce9592365225b93d8b28/utils/eval_with_pngs.py) in the root folder
+* `python run_monodepth.py --model_type dpt_hybrid_nyu --absolute_depth`
+(or **for new model** `python run_monodepth.py --model_type dpt_hybrid_nyu --absolute_depth --model_weights weights/dpt_hybrid_nyu_new-217f207d.pt` )
+* `python ./eval_with_pngs.py --pred_path ./output_monodepth/ --gt_path ./input/gt/ --dataset nyu --max_depth_eval 10  --eigen_crop`
+
+Result (old model) - **from paper**:
+```
+Evaluating 654 files
+GT files reading done
+0 GT files missing
+Computing errors
+     d1,      d2,      d3,  AbsRel,   SqRel,    RMSE, RMSElog,   SILog,   log10
+  0.904,   0.988,   0.998,   0.109,   0.054,   0.357,   0.129,   9.521,   0.045
+Done.
+```
+
+Result (new model):
+```
+GT files reading done
+697 GT files missing
+Computing errors
+     d1,      d2,      d3,  AbsRel,   SqRel,    RMSE, RMSElog,   SILog,   log10
+  0.905,   0.988,   0.998,   0.109,   0.055,   0.357,   0.129,   9.427,   0.045
+Done.
+```

DPT/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2021 Intel ISL (Intel Intelligent Systems Lab)
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

DPT/README.md

@@ -0,0 +1,97 @@
+## Vision Transformers for Dense Prediction
+
+This repository contains code and models for our [paper](https://arxiv.org/abs/2103.13413):
+
+> Vision Transformers for Dense Prediction  
+> René Ranftl, Alexey Bochkovskiy, Vladlen Koltun
+
+
+### Changelog 
+* [March 2021] Initial release of inference code and models
+
+### Setup 
+
+1) Download the model weights and place them in the `weights` folder:
+
+
+Monodepth:
+- [dpt_hybrid-midas-501f0c75.pt](https://github.com/intel-isl/DPT/releases/download/1_0/dpt_hybrid-midas-501f0c75.pt), [Mirror](https://drive.google.com/file/d/1dgcJEYYw1F8qirXhZxgNK8dWWz_8gZBD/view?usp=sharing)
+- [dpt_large-midas-2f21e586.pt](https://github.com/intel-isl/DPT/releases/download/1_0/dpt_large-midas-2f21e586.pt), [Mirror](https://drive.google.com/file/d/1vnuhoMc6caF-buQQ4hK0CeiMk9SjwB-G/view?usp=sharing)
+
+Segmentation:
+ - [dpt_hybrid-ade20k-53898607.pt](https://github.com/intel-isl/DPT/releases/download/1_0/dpt_hybrid-ade20k-53898607.pt), [Mirror](https://drive.google.com/file/d/1zKIAMbltJ3kpGLMh6wjsq65_k5XQ7_9m/view?usp=sharing)
+ - [dpt_large-ade20k-b12dca68.pt](https://github.com/intel-isl/DPT/releases/download/1_0/dpt_large-ade20k-b12dca68.pt), [Mirror](https://drive.google.com/file/d/1foDpUM7CdS8Zl6GPdkrJaAOjskb7hHe-/view?usp=sharing)
+  
+2) Set up dependencies: 
+
+    ```shell
+    pip install -r requirements.txt
+    ```
+
+   The code was tested with Python 3.7, PyTorch 1.8.0, OpenCV 4.5.1, and timm 0.4.5
+
+### Usage 
+
+1) Place one or more input images in the folder `input`.
+
+2) Run a monocular depth estimation model:
+
+    ```shell
+    python run_monodepth.py
+    ```
+
+    Or run a semantic segmentation model:
+
+    ```shell
+    python run_segmentation.py
+    ```
+
+3) The results are written to the folder `output_monodepth` and `output_semseg`, respectively.
+
+Use the flag `-t` to switch between different models. Possible options are `dpt_hybrid` (default) and `dpt_large`.
+
+
+**Additional models:**
+
+- Monodepth finetuned on KITTI: [dpt_hybrid_kitti-cb926ef4.pt](https://github.com/intel-isl/DPT/releases/download/1_0/dpt_hybrid_kitti-cb926ef4.pt) [Mirror](https://drive.google.com/file/d/1-oJpORoJEdxj4LTV-Pc17iB-smp-khcX/view?usp=sharing)
+- Monodepth finetuned on NYUv2: [dpt_hybrid_nyu-2ce69ec7.pt](https://github.com/intel-isl/DPT/releases/download/1_0/dpt_hybrid_nyu-2ce69ec7.pt) [Mirror](https\://drive.google.com/file/d/1NjiFw1Z9lUAfTPZu4uQ9gourVwvmd58O/view?usp=sharing)
+
+Run with 
+
+```shell
+python run_monodepth -t [dpt_hybrid_kitti|dpt_hybrid_nyu] 
+```
+
+### Evaluation
+
+Hints on how to evaluate monodepth models can be found here: https://github.com/intel-isl/DPT/blob/main/EVALUATION.md
+
+
+### Citation
+
+Please cite our papers if you use this code or any of the models. 
+```
+@article{Ranftl2021,
+	author    = {Ren\'{e} Ranftl and Alexey Bochkovskiy and Vladlen Koltun},
+	title     = {Vision Transformers for Dense Prediction},
+	journal   = {ArXiv preprint},
+	year      = {2021},
+}
+```
+
+```
+@article{Ranftl2020,
+	author    = {Ren\'{e} Ranftl and Katrin Lasinger and David Hafner and Konrad Schindler and Vladlen Koltun},
+	title     = {Towards Robust Monocular Depth Estimation: Mixing Datasets for Zero-shot Cross-dataset Transfer},
+	journal   = {IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI)},
+	year      = {2020},
+}
+```
+
+### Acknowledgements
+
+Our work builds on and uses code from [timm](https://github.com/rwightman/pytorch-image-models) and [PyTorch-Encoding](https://github.com/zhanghang1989/PyTorch-Encoding). We'd like to thank the authors for making these libraries available.
+
+### License 
+
+MIT License 

DPT/dpt/base_model.py

@@ -0,0 +1,16 @@
+import torch
+
+
+class BaseModel(torch.nn.Module):
+    def load(self, path):
+        """Load model from file.
+
+        Args:
+            path (str): file path
+        """
+        parameters = torch.load(path, map_location=torch.device("cpu"))
+
+        if "optimizer" in parameters:
+            parameters = parameters["model"]
+
+        self.load_state_dict(parameters)

DPT/dpt/blocks.py

@@ -0,0 +1,383 @@
+import torch
+import torch.nn as nn
+
+from .vit import (
+    _make_pretrained_vitb_rn50_384,
+    _make_pretrained_vitl16_384,
+    _make_pretrained_vitb16_384,
+    forward_vit,
+)
+
+
+def _make_encoder(
+    backbone,
+    features,
+    use_pretrained,
+    groups=1,
+    expand=False,
+    exportable=True,
+    hooks=None,
+    use_vit_only=False,
+    use_readout="ignore",
+    enable_attention_hooks=False,
+):
+    if backbone == "vitl16_384":
+        pretrained = _make_pretrained_vitl16_384(
+            use_pretrained,
+            hooks=hooks,
+            use_readout=use_readout,
+            enable_attention_hooks=enable_attention_hooks,
+        )
+        scratch = _make_scratch(
+            [256, 512, 1024, 1024], features, groups=groups, expand=expand
+        )  # ViT-L/16 - 85.0% Top1 (backbone)
+    elif backbone == "vitb_rn50_384":
+        pretrained = _make_pretrained_vitb_rn50_384(
+            use_pretrained,
+            hooks=hooks,
+            use_vit_only=use_vit_only,
+            use_readout=use_readout,
+            enable_attention_hooks=enable_attention_hooks,
+        )
+        scratch = _make_scratch(
+            [256, 512, 768, 768], features, groups=groups, expand=expand
+        )  # ViT-H/16 - 85.0% Top1 (backbone)
+    elif backbone == "vitb16_384":
+        pretrained = _make_pretrained_vitb16_384(
+            use_pretrained,
+            hooks=hooks,
+            use_readout=use_readout,
+            enable_attention_hooks=enable_attention_hooks,
+        )
+        scratch = _make_scratch(
+            [96, 192, 384, 768], features, groups=groups, expand=expand
+        )  # ViT-B/16 - 84.6% Top1 (backbone)
+    elif backbone == "resnext101_wsl":
+        pretrained = _make_pretrained_resnext101_wsl(use_pretrained)
+        scratch = _make_scratch(
+            [256, 512, 1024, 2048], features, groups=groups, expand=expand
+        )  # efficientnet_lite3
+    else:
+        print(f"Backbone '{backbone}' not implemented")
+        assert False
+
+    return pretrained, scratch
+
+
+def _make_scratch(in_shape, out_shape, groups=1, expand=False):
+    scratch = nn.Module()
+
+    out_shape1 = out_shape
+    out_shape2 = out_shape
+    out_shape3 = out_shape
+    out_shape4 = out_shape
+    if expand == True:
+        out_shape1 = out_shape
+        out_shape2 = out_shape * 2
+        out_shape3 = out_shape * 4
+        out_shape4 = out_shape * 8
+
+    scratch.layer1_rn = nn.Conv2d(
+        in_shape[0],
+        out_shape1,
+        kernel_size=3,
+        stride=1,
+        padding=1,
+        bias=False,
+        groups=groups,
+    )
+    scratch.layer2_rn = nn.Conv2d(
+        in_shape[1],
+        out_shape2,
+        kernel_size=3,
+        stride=1,
+        padding=1,
+        bias=False,
+        groups=groups,
+    )
+    scratch.layer3_rn = nn.Conv2d(
+        in_shape[2],
+        out_shape3,
+        kernel_size=3,
+        stride=1,
+        padding=1,
+        bias=False,
+        groups=groups,
+    )
+    scratch.layer4_rn = nn.Conv2d(
+        in_shape[3],
+        out_shape4,
+        kernel_size=3,
+        stride=1,
+        padding=1,
+        bias=False,
+        groups=groups,
+    )
+
+    return scratch
+
+
+def _make_resnet_backbone(resnet):
+    pretrained = nn.Module()
+    pretrained.layer1 = nn.Sequential(
+        resnet.conv1, resnet.bn1, resnet.relu, resnet.maxpool, resnet.layer1
+    )
+
+    pretrained.layer2 = resnet.layer2
+    pretrained.layer3 = resnet.layer3
+    pretrained.layer4 = resnet.layer4
+
+    return pretrained
+
+
+def _make_pretrained_resnext101_wsl(use_pretrained):
+    resnet = torch.hub.load("facebookresearch/WSL-Images", "resnext101_32x8d_wsl")
+    return _make_resnet_backbone(resnet)
+
+
+class Interpolate(nn.Module):
+    """Interpolation module."""
+
+    def __init__(self, scale_factor, mode, align_corners=False):
+        """Init.
+
+        Args:
+            scale_factor (float): scaling
+            mode (str): interpolation mode
+        """
+        super(Interpolate, self).__init__()
+
+        self.interp = nn.functional.interpolate
+        self.scale_factor = scale_factor
+        self.mode = mode
+        self.align_corners = align_corners
+
+    def forward(self, x):
+        """Forward pass.
+
+        Args:
+            x (tensor): input
+
+        Returns:
+            tensor: interpolated data
+        """
+
+        x = self.interp(
+            x,
+            scale_factor=self.scale_factor,
+            mode=self.mode,
+            align_corners=self.align_corners,
+        )
+
+        return x
+
+
+class ResidualConvUnit(nn.Module):
+    """Residual convolution module."""
+
+    def __init__(self, features):
+        """Init.
+
+        Args:
+            features (int): number of features
+        """
+        super().__init__()
+
+        self.conv1 = nn.Conv2d(
+            features, features, kernel_size=3, stride=1, padding=1, bias=True
+        )
+
+        self.conv2 = nn.Conv2d(
+            features, features, kernel_size=3, stride=1, padding=1, bias=True
+        )
+
+        self.relu = nn.ReLU(inplace=True)
+
+    def forward(self, x):
+        """Forward pass.
+
+        Args:
+            x (tensor): input
+
+        Returns:
+            tensor: output
+        """
+        out = self.relu(x)
+        out = self.conv1(out)
+        out = self.relu(out)
+        out = self.conv2(out)
+
+        return out + x
+
+
+class FeatureFusionBlock(nn.Module):
+    """Feature fusion block."""
+
+    def __init__(self, features):
+        """Init.
+
+        Args:
+            features (int): number of features
+        """
+        super(FeatureFusionBlock, self).__init__()
+
+        self.resConfUnit1 = ResidualConvUnit(features)
+        self.resConfUnit2 = ResidualConvUnit(features)
+
+    def forward(self, *xs):
+        """Forward pass.
+
+        Returns:
+            tensor: output
+        """
+        output = xs[0]
+
+        if len(xs) == 2:
+            output += self.resConfUnit1(xs[1])
+
+        output = self.resConfUnit2(output)
+
+        output = nn.functional.interpolate(
+            output, scale_factor=2, mode="bilinear", align_corners=True
+        )
+
+        return output
+
+
+class ResidualConvUnit_custom(nn.Module):
+    """Residual convolution module."""
+
+    def __init__(self, features, activation, bn):
+        """Init.
+
+        Args:
+            features (int): number of features
+        """
+        super().__init__()
+
+        self.bn = bn
+
+        self.groups = 1
+
+        self.conv1 = nn.Conv2d(
+            features,
+            features,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            bias=not self.bn,
+            groups=self.groups,
+        )
+
+        self.conv2 = nn.Conv2d(
+            features,
+            features,
+            kernel_size=3,
+            stride=1,
+            padding=1,
+            bias=not self.bn,
+            groups=self.groups,
+        )
+
+        if self.bn == True:
+            self.bn1 = nn.BatchNorm2d(features)
+            self.bn2 = nn.BatchNorm2d(features)
+
+        self.activation = activation
+
+        self.skip_add = nn.quantized.FloatFunctional()
+
+    def forward(self, x):
+        """Forward pass.
+
+        Args:
+            x (tensor): input
+
+        Returns:
+            tensor: output
+        """
+
+        out = self.activation(x)
+        out = self.conv1(out)
+        if self.bn == True:
+            out = self.bn1(out)
+
+        out = self.activation(out)
+        out = self.conv2(out)
+        if self.bn == True:
+            out = self.bn2(out)
+
+        if self.groups > 1:
+            out = self.conv_merge(out)
+
+        return self.skip_add.add(out, x)
+
+        # return out + x
+
+
+class FeatureFusionBlock_custom(nn.Module):
+    """Feature fusion block."""
+
+    def __init__(
+        self,
+        features,
+        activation,
+        deconv=False,
+        bn=False,
+        expand=False,
+        align_corners=True,
+    ):
+        """Init.
+
+        Args:
+            features (int): number of features
+        """
+        super(FeatureFusionBlock_custom, self).__init__()
+
+        self.deconv = deconv
+        self.align_corners = align_corners
+
+        self.groups = 1
+
+        self.expand = expand
+        out_features = features
+        if self.expand == True:
+            out_features = features // 2
+
+        self.out_conv = nn.Conv2d(
+            features,
+            out_features,
+            kernel_size=1,
+            stride=1,
+            padding=0,
+            bias=True,
+            groups=1,
+        )
+
+        self.resConfUnit1 = ResidualConvUnit_custom(features, activation, bn)
+        self.resConfUnit2 = ResidualConvUnit_custom(features, activation, bn)
+
+        self.skip_add = nn.quantized.FloatFunctional()
+
+    def forward(self, *xs):
+        """Forward pass.
+
+        Returns:
+            tensor: output
+        """
+        output = xs[0]
+
+        if len(xs) == 2:
+            res = self.resConfUnit1(xs[1])
+            output = self.skip_add.add(output, res)
+            # output += res
+
+        output = self.resConfUnit2(output)
+
+        output = nn.functional.interpolate(
+            output, scale_factor=2, mode="bilinear", align_corners=self.align_corners
+        )
+
+        output = self.out_conv(output)
+
+        return output

DPT/dpt/midas_net.py

@@ -0,0 +1,77 @@
+"""MidashNet: Network for monocular depth estimation trained by mixing several datasets.
+This file contains code that is adapted from
+https://github.com/thomasjpfan/pytorch_refinenet/blob/master/pytorch_refinenet/refinenet/refinenet_4cascade.py
+"""
+import torch
+import torch.nn as nn
+
+from .base_model import BaseModel
+from .blocks import FeatureFusionBlock, Interpolate, _make_encoder
+
+
+class MidasNet_large(BaseModel):
+    """Network for monocular depth estimation."""
+
+    def __init__(self, path=None, features=256, non_negative=True):
+        """Init.
+
+        Args:
+            path (str, optional): Path to saved model. Defaults to None.
+            features (int, optional): Number of features. Defaults to 256.
+            backbone (str, optional): Backbone network for encoder. Defaults to resnet50
+        """
+        print("Loading weights: ", path)
+
+        super(MidasNet_large, self).__init__()
+
+        use_pretrained = False if path is None else True
+
+        self.pretrained, self.scratch = _make_encoder(
+            backbone="resnext101_wsl", features=features, use_pretrained=use_pretrained
+        )
+
+        self.scratch.refinenet4 = FeatureFusionBlock(features)
+        self.scratch.refinenet3 = FeatureFusionBlock(features)
+        self.scratch.refinenet2 = FeatureFusionBlock(features)
+        self.scratch.refinenet1 = FeatureFusionBlock(features)
+
+        self.scratch.output_conv = nn.Sequential(
+            nn.Conv2d(features, 128, kernel_size=3, stride=1, padding=1),
+            Interpolate(scale_factor=2, mode="bilinear"),
+            nn.Conv2d(128, 32, kernel_size=3, stride=1, padding=1),
+            nn.ReLU(True),
+            nn.Conv2d(32, 1, kernel_size=1, stride=1, padding=0),
+            nn.ReLU(True) if non_negative else nn.Identity(),
+        )
+
+        if path:
+            self.load(path)
+
+    def forward(self, x):
+        """Forward pass.
+
+        Args:
+            x (tensor): input data (image)
+
+        Returns:
+            tensor: depth
+        """
+
+        layer_1 = self.pretrained.layer1(x)
+        layer_2 = self.pretrained.layer2(layer_1)
+        layer_3 = self.pretrained.layer3(layer_2)
+        layer_4 = self.pretrained.layer4(layer_3)
+
+        layer_1_rn = self.scratch.layer1_rn(layer_1)
+        layer_2_rn = self.scratch.layer2_rn(layer_2)
+        layer_3_rn = self.scratch.layer3_rn(layer_3)
+        layer_4_rn = self.scratch.layer4_rn(layer_4)
+
+        path_4 = self.scratch.refinenet4(layer_4_rn)
+        path_3 = self.scratch.refinenet3(path_4, layer_3_rn)
+        path_2 = self.scratch.refinenet2(path_3, layer_2_rn)
+        path_1 = self.scratch.refinenet1(path_2, layer_1_rn)
+
+        out = self.scratch.output_conv(path_1)
+
+        return torch.squeeze(out, dim=1)

DPT/dpt/models.py

@@ -0,0 +1,153 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from .base_model import BaseModel
+from .blocks import (
+    FeatureFusionBlock,
+    FeatureFusionBlock_custom,
+    Interpolate,
+    _make_encoder,
+    forward_vit,
+)
+
+
+def _make_fusion_block(features, use_bn):
+    return FeatureFusionBlock_custom(
+        features,
+        nn.ReLU(False),
+        deconv=False,
+        bn=use_bn,
+        expand=False,
+        align_corners=True,
+    )
+
+
+class DPT(BaseModel):
+    def __init__(
+        self,
+        head,
+        features=256,
+        backbone="vitb_rn50_384",
+        readout="project",
+        channels_last=False,
+        use_bn=False,
+        enable_attention_hooks=False,
+    ):
+
+        super(DPT, self).__init__()
+
+        self.channels_last = channels_last
+
+        hooks = {
+            "vitb_rn50_384": [0, 1, 8, 11],
+            "vitb16_384": [2, 5, 8, 11],
+            "vitl16_384": [5, 11, 17, 23],
+        }
+
+        # Instantiate backbone and reassemble blocks
+        self.pretrained, self.scratch = _make_encoder(
+            backbone,
+            features,
+            False,  # Set to true of you want to train from scratch, uses ImageNet weights
+            groups=1,
+            expand=False,
+            exportable=False,
+            hooks=hooks[backbone],
+            use_readout=readout,
+            enable_attention_hooks=enable_attention_hooks,
+        )
+
+        self.scratch.refinenet1 = _make_fusion_block(features, use_bn)
+        self.scratch.refinenet2 = _make_fusion_block(features, use_bn)
+        self.scratch.refinenet3 = _make_fusion_block(features, use_bn)
+        self.scratch.refinenet4 = _make_fusion_block(features, use_bn)
+
+        self.scratch.output_conv = head
+
+    def forward(self, x):
+        if self.channels_last == True:
+            x.contiguous(memory_format=torch.channels_last)
+
+        layer_1, layer_2, layer_3, layer_4 = forward_vit(self.pretrained, x)
+
+        layer_1_rn = self.scratch.layer1_rn(layer_1)
+        layer_2_rn = self.scratch.layer2_rn(layer_2)
+        layer_3_rn = self.scratch.layer3_rn(layer_3)
+        layer_4_rn = self.scratch.layer4_rn(layer_4)
+
+        path_4 = self.scratch.refinenet4(layer_4_rn)
+        path_3 = self.scratch.refinenet3(path_4, layer_3_rn)
+        path_2 = self.scratch.refinenet2(path_3, layer_2_rn)
+        path_1 = self.scratch.refinenet1(path_2, layer_1_rn)
+
+        out = self.scratch.output_conv(path_1)
+
+        return out
+
+
+class DPTDepthModel(DPT):
+    def __init__(
+        self, path=None, non_negative=True, scale=1.0, shift=0.0, invert=False, **kwargs
+    ):
+        features = kwargs["features"] if "features" in kwargs else 256
+
+        self.scale = scale
+        self.shift = shift
+        self.invert = invert
+
+        head = nn.Sequential(
+            nn.Conv2d(features, features // 2, kernel_size=3, stride=1, padding=1),
+            Interpolate(scale_factor=2, mode="bilinear", align_corners=True),
+            nn.Conv2d(features // 2, 32, kernel_size=3, stride=1, padding=1),
+            nn.ReLU(True),
+            nn.Conv2d(32, 1, kernel_size=1, stride=1, padding=0),
+            nn.ReLU(True) if non_negative else nn.Identity(),
+            nn.Identity(),
+        )
+
+        super().__init__(head, **kwargs)
+
+        if path is not None:
+            self.load(path)
+
+    def forward(self, x):
+        inv_depth = super().forward(x).squeeze(dim=1)
+
+        if self.invert:
+            depth = self.scale * inv_depth + self.shift
+            depth[depth < 1e-8] = 1e-8
+            depth = 1.0 / depth
+            return depth
+        else:
+            return inv_depth
+
+
+class DPTSegmentationModel(DPT):
+    def __init__(self, num_classes, path=None, **kwargs):
+
+        features = kwargs["features"] if "features" in kwargs else 256
+
+        kwargs["use_bn"] = True
+
+        head = nn.Sequential(
+            nn.Conv2d(features, features, kernel_size=3, padding=1, bias=False),
+            nn.BatchNorm2d(features),
+            nn.ReLU(True),
+            nn.Dropout(0.1, False),
+            nn.Conv2d(features, num_classes, kernel_size=1),
+            Interpolate(scale_factor=2, mode="bilinear", align_corners=True),
+        )
+
+        super().__init__(head, **kwargs)
+
+        self.auxlayer = nn.Sequential(
+            nn.Conv2d(features, features, kernel_size=3, padding=1, bias=False),
+            nn.BatchNorm2d(features),
+            nn.ReLU(True),
+            nn.Dropout(0.1, False),
+            nn.Conv2d(features, num_classes, kernel_size=1),
+        )
+
+        if path is not None:
+            self.load(path)

DPT/dpt/transforms.py

@@ -0,0 +1,231 @@
+import numpy as np
+import cv2
+import math
+
+
+def apply_min_size(sample, size, image_interpolation_method=cv2.INTER_AREA):
+    """Rezise the sample to ensure the given size. Keeps aspect ratio.
+
+    Args:
+        sample (dict): sample
+        size (tuple): image size
+
+    Returns:
+        tuple: new size
+    """
+    shape = list(sample["disparity"].shape)
+
+    if shape[0] >= size[0] and shape[1] >= size[1]:
+        return sample
+
+    scale = [0, 0]
+    scale[0] = size[0] / shape[0]
+    scale[1] = size[1] / shape[1]
+
+    scale = max(scale)
+
+    shape[0] = math.ceil(scale * shape[0])
+    shape[1] = math.ceil(scale * shape[1])
+
+    # resize
+    sample["image"] = cv2.resize(
+        sample["image"], tuple(shape[::-1]), interpolation=image_interpolation_method
+    )
+
+    sample["disparity"] = cv2.resize(
+        sample["disparity"], tuple(shape[::-1]), interpolation=cv2.INTER_NEAREST
+    )
+    sample["mask"] = cv2.resize(
+        sample["mask"].astype(np.float32),
+        tuple(shape[::-1]),
+        interpolation=cv2.INTER_NEAREST,
+    )
+    sample["mask"] = sample["mask"].astype(bool)
+
+    return tuple(shape)
+
+
+class Resize(object):
+    """Resize sample to given size (width, height)."""
+
+    def __init__(
+        self,
+        width,
+        height,
+        resize_target=True,
+        keep_aspect_ratio=False,
+        ensure_multiple_of=1,
+        resize_method="lower_bound",
+        image_interpolation_method=cv2.INTER_AREA,
+    ):
+        """Init.
+
+        Args:
+            width (int): desired output width
+            height (int): desired output height
+            resize_target (bool, optional):
+                True: Resize the full sample (image, mask, target).
+                False: Resize image only.
+                Defaults to True.
+            keep_aspect_ratio (bool, optional):
+                True: Keep the aspect ratio of the input sample.
+                Output sample might not have the given width and height, and
+                resize behaviour depends on the parameter 'resize_method'.
+                Defaults to False.
+            ensure_multiple_of (int, optional):
+                Output width and height is constrained to be multiple of this parameter.
+                Defaults to 1.
+            resize_method (str, optional):
+                "lower_bound": Output will be at least as large as the given size.
+                "upper_bound": Output will be at max as large as the given size. (Output size might be smaller than given size.)
+                "minimal": Scale as least as possible.  (Output size might be smaller than given size.)
+                Defaults to "lower_bound".
+        """
+        self.__width = width
+        self.__height = height
+
+        self.__resize_target = resize_target
+        self.__keep_aspect_ratio = keep_aspect_ratio
+        self.__multiple_of = ensure_multiple_of
+        self.__resize_method = resize_method
+        self.__image_interpolation_method = image_interpolation_method
+
+    def constrain_to_multiple_of(self, x, min_val=0, max_val=None):
+        y = (np.round(x / self.__multiple_of) * self.__multiple_of).astype(int)
+
+        if max_val is not None and y > max_val:
+            y = (np.floor(x / self.__multiple_of) * self.__multiple_of).astype(int)
+
+        if y < min_val:
+            y = (np.ceil(x / self.__multiple_of) * self.__multiple_of).astype(int)
+
+        return y
+
+    def get_size(self, width, height):
+        # determine new height and width
+        scale_height = self.__height / height
+        scale_width = self.__width / width
+
+        if self.__keep_aspect_ratio:
+            if self.__resize_method == "lower_bound":
+                # scale such that output size is lower bound
+                if scale_width > scale_height:
+                    # fit width
+                    scale_height = scale_width
+                else:
+                    # fit height
+                    scale_width = scale_height
+            elif self.__resize_method == "upper_bound":
+                # scale such that output size is upper bound
+                if scale_width < scale_height:
+                    # fit width
+                    scale_height = scale_width
+                else:
+                    # fit height
+                    scale_width = scale_height
+            elif self.__resize_method == "minimal":
+                # scale as least as possbile
+                if abs(1 - scale_width) < abs(1 - scale_height):
+                    # fit width
+                    scale_height = scale_width
+                else:
+                    # fit height
+                    scale_width = scale_height
+            else:
+                raise ValueError(
+                    f"resize_method {self.__resize_method} not implemented"
+                )
+
+        if self.__resize_method == "lower_bound":
+            new_height = self.constrain_to_multiple_of(
+                scale_height * height, min_val=self.__height
+            )
+            new_width = self.constrain_to_multiple_of(
+                scale_width * width, min_val=self.__width
+            )
+        elif self.__resize_method == "upper_bound":
+            new_height = self.constrain_to_multiple_of(
+                scale_height * height, max_val=self.__height
+            )
+            new_width = self.constrain_to_multiple_of(
+                scale_width * width, max_val=self.__width
+            )
+        elif self.__resize_method == "minimal":
+            new_height = self.constrain_to_multiple_of(scale_height * height)
+            new_width = self.constrain_to_multiple_of(scale_width * width)
+        else:
+            raise ValueError(f"resize_method {self.__resize_method} not implemented")
+
+        return (new_width, new_height)
+
+    def __call__(self, sample):
+        width, height = self.get_size(
+            sample["image"].shape[1], sample["image"].shape[0]
+        )
+
+        # resize sample
+        sample["image"] = cv2.resize(
+            sample["image"],
+            (width, height),
+            interpolation=self.__image_interpolation_method,
+        )
+
+        if self.__resize_target:
+            if "disparity" in sample:
+                sample["disparity"] = cv2.resize(
+                    sample["disparity"],
+                    (width, height),
+                    interpolation=cv2.INTER_NEAREST,
+                )
+
+            if "depth" in sample:
+                sample["depth"] = cv2.resize(
+                    sample["depth"], (width, height), interpolation=cv2.INTER_NEAREST
+                )
+
+            sample["mask"] = cv2.resize(
+                sample["mask"].astype(np.float32),
+                (width, height),
+                interpolation=cv2.INTER_NEAREST,
+            )
+            sample["mask"] = sample["mask"].astype(bool)
+
+        return sample
+
+
+class NormalizeImage(object):
+    """Normlize image by given mean and std."""
+
+    def __init__(self, mean, std):
+        self.__mean = mean
+        self.__std = std
+
+    def __call__(self, sample):
+        sample["image"] = (sample["image"] - self.__mean) / self.__std
+
+        return sample
+
+
+class PrepareForNet(object):
+    """Prepare sample for usage as network input."""
+
+    def __init__(self):
+        pass
+
+    def __call__(self, sample):
+        image = np.transpose(sample["image"], (2, 0, 1))
+        sample["image"] = np.ascontiguousarray(image).astype(np.float32)
+
+        if "mask" in sample:
+            sample["mask"] = sample["mask"].astype(np.float32)
+            sample["mask"] = np.ascontiguousarray(sample["mask"])
+
+        if "disparity" in sample:
+            disparity = sample["disparity"].astype(np.float32)
+            sample["disparity"] = np.ascontiguousarray(disparity)
+
+        if "depth" in sample:
+            depth = sample["depth"].astype(np.float32)
+            sample["depth"] = np.ascontiguousarray(depth)
+
+        return sample

DPT/dpt/vit.py

@@ -0,0 +1,576 @@
+import torch
+import torch.nn as nn
+import timm
+import types
+import math
+import torch.nn.functional as F
+
+
+activations = {}
+
+
+def get_activation(name):
+    def hook(model, input, output):
+        activations[name] = output
+
+    return hook
+
+
+attention = {}
+
+
+def get_attention(name):
+    def hook(module, input, output):
+        x = input[0]
+        B, N, C = x.shape
+        qkv = (
+            module.qkv(x)
+            .reshape(B, N, 3, module.num_heads, C // module.num_heads)
+            .permute(2, 0, 3, 1, 4)
+        )
+        q, k, v = (
+            qkv[0],
+            qkv[1],
+            qkv[2],
+        )  # make torchscript happy (cannot use tensor as tuple)
+
+        attn = (q @ k.transpose(-2, -1)) * module.scale
+
+        attn = attn.softmax(dim=-1)  # [:,:,1,1:]
+        attention[name] = attn
+
+    return hook
+
+
+def get_mean_attention_map(attn, token, shape):
+    attn = attn[:, :, token, 1:]
+    attn = attn.unflatten(2, torch.Size([shape[2] // 16, shape[3] // 16])).float()
+    attn = torch.nn.functional.interpolate(
+        attn, size=shape[2:], mode="bicubic", align_corners=False
+    ).squeeze(0)
+
+    all_attn = torch.mean(attn, 0)
+
+    return all_attn
+
+
+class Slice(nn.Module):
+    def __init__(self, start_index=1):
+        super(Slice, self).__init__()
+        self.start_index = start_index
+
+    def forward(self, x):
+        return x[:, self.start_index :]
+
+
+class AddReadout(nn.Module):
+    def __init__(self, start_index=1):
+        super(AddReadout, self).__init__()
+        self.start_index = start_index
+
+    def forward(self, x):
+        if self.start_index == 2:
+            readout = (x[:, 0] + x[:, 1]) / 2
+        else:
+            readout = x[:, 0]
+        return x[:, self.start_index :] + readout.unsqueeze(1)
+
+
+class ProjectReadout(nn.Module):
+    def __init__(self, in_features, start_index=1):
+        super(ProjectReadout, self).__init__()
+        self.start_index = start_index
+
+        self.project = nn.Sequential(nn.Linear(2 * in_features, in_features), nn.GELU())
+
+    def forward(self, x):
+        readout = x[:, 0].unsqueeze(1).expand_as(x[:, self.start_index :])
+        features = torch.cat((x[:, self.start_index :], readout), -1)
+
+        return self.project(features)
+
+
+class Transpose(nn.Module):
+    def __init__(self, dim0, dim1):
+        super(Transpose, self).__init__()
+        self.dim0 = dim0
+        self.dim1 = dim1
+
+    def forward(self, x):
+        x = x.transpose(self.dim0, self.dim1)
+        return x
+
+
+def forward_vit(pretrained, x):
+    b, c, h, w = x.shape
+
+    glob = pretrained.model.forward_flex(x)
+
+    layer_1 = pretrained.activations["1"]
+    layer_2 = pretrained.activations["2"]
+    layer_3 = pretrained.activations["3"]
+    layer_4 = pretrained.activations["4"]
+
+    layer_1 = pretrained.act_postprocess1[0:2](layer_1)
+    layer_2 = pretrained.act_postprocess2[0:2](layer_2)
+    layer_3 = pretrained.act_postprocess3[0:2](layer_3)
+    layer_4 = pretrained.act_postprocess4[0:2](layer_4)
+
+    unflatten = nn.Sequential(
+        nn.Unflatten(
+            2,
+            torch.Size(
+                [
+                    h // pretrained.model.patch_size[1],
+                    w // pretrained.model.patch_size[0],
+                ]
+            ),
+        )
+    )
+
+    if layer_1.ndim == 3:
+        layer_1 = unflatten(layer_1)
+    if layer_2.ndim == 3:
+        layer_2 = unflatten(layer_2)
+    if layer_3.ndim == 3:
+        layer_3 = unflatten(layer_3)
+    if layer_4.ndim == 3:
+        layer_4 = unflatten(layer_4)
+
+    layer_1 = pretrained.act_postprocess1[3 : len(pretrained.act_postprocess1)](layer_1)
+    layer_2 = pretrained.act_postprocess2[3 : len(pretrained.act_postprocess2)](layer_2)
+    layer_3 = pretrained.act_postprocess3[3 : len(pretrained.act_postprocess3)](layer_3)
+    layer_4 = pretrained.act_postprocess4[3 : len(pretrained.act_postprocess4)](layer_4)
+
+    return layer_1, layer_2, layer_3, layer_4
+
+
+def _resize_pos_embed(self, posemb, gs_h, gs_w):
+    posemb_tok, posemb_grid = (
+        posemb[:, : self.start_index],
+        posemb[0, self.start_index :],
+    )
+
+    gs_old = int(math.sqrt(len(posemb_grid)))
+
+    posemb_grid = posemb_grid.reshape(1, gs_old, gs_old, -1).permute(0, 3, 1, 2)
+    posemb_grid = F.interpolate(posemb_grid, size=(gs_h, gs_w), mode="bilinear")
+    posemb_grid = posemb_grid.permute(0, 2, 3, 1).reshape(1, gs_h * gs_w, -1)
+
+    posemb = torch.cat([posemb_tok, posemb_grid], dim=1)
+
+    return posemb
+
+
+def forward_flex(self, x):
+    b, c, h, w = x.shape
+
+    pos_embed = self._resize_pos_embed(
+        self.pos_embed, h // self.patch_size[1], w // self.patch_size[0]
+    )
+
+    B = x.shape[0]
+
+    if hasattr(self.patch_embed, "backbone"):
+        x = self.patch_embed.backbone(x)
+        if isinstance(x, (list, tuple)):
+            x = x[-1]  # last feature if backbone outputs list/tuple of features
+
+    x = self.patch_embed.proj(x).flatten(2).transpose(1, 2)
+
+    if getattr(self, "dist_token", None) is not None:
+        cls_tokens = self.cls_token.expand(
+            B, -1, -1
+        )  # stole cls_tokens impl from Phil Wang, thanks
+        dist_token = self.dist_token.expand(B, -1, -1)
+        x = torch.cat((cls_tokens, dist_token, x), dim=1)
+    else:
+        cls_tokens = self.cls_token.expand(
+            B, -1, -1
+        )  # stole cls_tokens impl from Phil Wang, thanks
+        x = torch.cat((cls_tokens, x), dim=1)
+
+    x = x + pos_embed
+    x = self.pos_drop(x)
+
+    for blk in self.blocks:
+        x = blk(x)
+
+    x = self.norm(x)
+
+    return x
+
+
+def get_readout_oper(vit_features, features, use_readout, start_index=1):
+    if use_readout == "ignore":
+        readout_oper = [Slice(start_index)] * len(features)
+    elif use_readout == "add":
+        readout_oper = [AddReadout(start_index)] * len(features)
+    elif use_readout == "project":
+        readout_oper = [
+            ProjectReadout(vit_features, start_index) for out_feat in features
+        ]
+    else:
+        assert (
+            False
+        ), "wrong operation for readout token, use_readout can be 'ignore', 'add', or 'project'"
+
+    return readout_oper
+
+
+def _make_vit_b16_backbone(
+    model,
+    features=[96, 192, 384, 768],
+    size=[384, 384],
+    hooks=[2, 5, 8, 11],
+    vit_features=768,
+    use_readout="ignore",
+    start_index=1,
+    enable_attention_hooks=False,
+):
+    pretrained = nn.Module()
+
+    pretrained.model = model
+    pretrained.model.blocks[hooks[0]].register_forward_hook(get_activation("1"))
+    pretrained.model.blocks[hooks[1]].register_forward_hook(get_activation("2"))
+    pretrained.model.blocks[hooks[2]].register_forward_hook(get_activation("3"))
+    pretrained.model.blocks[hooks[3]].register_forward_hook(get_activation("4"))
+
+    pretrained.activations = activations
+
+    if enable_attention_hooks:
+        pretrained.model.blocks[hooks[0]].attn.register_forward_hook(
+            get_attention("attn_1")
+        )
+        pretrained.model.blocks[hooks[1]].attn.register_forward_hook(
+            get_attention("attn_2")
+        )
+        pretrained.model.blocks[hooks[2]].attn.register_forward_hook(
+            get_attention("attn_3")
+        )
+        pretrained.model.blocks[hooks[3]].attn.register_forward_hook(
+            get_attention("attn_4")
+        )
+        pretrained.attention = attention
+
+    readout_oper = get_readout_oper(vit_features, features, use_readout, start_index)
+
+    # 32, 48, 136, 384
+    pretrained.act_postprocess1 = nn.Sequential(
+        readout_oper[0],
+        Transpose(1, 2),
+        nn.Unflatten(2, torch.Size([size[0] // 16, size[1] // 16])),
+        nn.Conv2d(
+            in_channels=vit_features,
+            out_channels=features[0],
+            kernel_size=1,
+            stride=1,
+            padding=0,
+        ),
+        nn.ConvTranspose2d(
+            in_channels=features[0],
+            out_channels=features[0],
+            kernel_size=4,
+            stride=4,
+            padding=0,
+            bias=True,
+            dilation=1,
+            groups=1,
+        ),
+    )
+
+    pretrained.act_postprocess2 = nn.Sequential(
+        readout_oper[1],
+        Transpose(1, 2),
+        nn.Unflatten(2, torch.Size([size[0] // 16, size[1] // 16])),
+        nn.Conv2d(
+            in_channels=vit_features,
+            out_channels=features[1],
+            kernel_size=1,
+            stride=1,
+            padding=0,
+        ),
+        nn.ConvTranspose2d(
+            in_channels=features[1],
+            out_channels=features[1],
+            kernel_size=2,
+            stride=2,
+            padding=0,
+            bias=True,
+            dilation=1,
+            groups=1,
+        ),
+    )
+
+    pretrained.act_postprocess3 = nn.Sequential(
+        readout_oper[2],
+        Transpose(1, 2),
+        nn.Unflatten(2, torch.Size([size[0] // 16, size[1] // 16])),
+        nn.Conv2d(
+            in_channels=vit_features,
+            out_channels=features[2],
+            kernel_size=1,
+            stride=1,
+            padding=0,
+        ),
+    )
+
+    pretrained.act_postprocess4 = nn.Sequential(
+        readout_oper[3],
+        Transpose(1, 2),
+        nn.Unflatten(2, torch.Size([size[0] // 16, size[1] // 16])),
+        nn.Conv2d(
+            in_channels=vit_features,
+            out_channels=features[3],
+            kernel_size=1,
+            stride=1,
+            padding=0,
+        ),
+        nn.Conv2d(
+            in_channels=features[3],
+            out_channels=features[3],
+            kernel_size=3,
+            stride=2,
+            padding=1,
+        ),
+    )
+
+    pretrained.model.start_index = start_index
+    pretrained.model.patch_size = [16, 16]
+
+    # We inject this function into the VisionTransformer instances so that
+    # we can use it with interpolated position embeddings without modifying the library source.
+    pretrained.model.forward_flex = types.MethodType(forward_flex, pretrained.model)
+    pretrained.model._resize_pos_embed = types.MethodType(
+        _resize_pos_embed, pretrained.model
+    )
+
+    return pretrained
+
+
+def _make_vit_b_rn50_backbone(
+    model,
+    features=[256, 512, 768, 768],
+    size=[384, 384],
+    hooks=[0, 1, 8, 11],
+    vit_features=768,
+    use_vit_only=False,
+    use_readout="ignore",
+    start_index=1,
+    enable_attention_hooks=False,
+):
+    pretrained = nn.Module()
+
+    pretrained.model = model
+
+    if use_vit_only == True:
+        pretrained.model.blocks[hooks[0]].register_forward_hook(get_activation("1"))
+        pretrained.model.blocks[hooks[1]].register_forward_hook(get_activation("2"))
+    else:
+        pretrained.model.patch_embed.backbone.stages[0].register_forward_hook(
+            get_activation("1")
+        )
+        pretrained.model.patch_embed.backbone.stages[1].register_forward_hook(
+            get_activation("2")
+        )
+
+    pretrained.model.blocks[hooks[2]].register_forward_hook(get_activation("3"))
+    pretrained.model.blocks[hooks[3]].register_forward_hook(get_activation("4"))
+
+    if enable_attention_hooks:
+        pretrained.model.blocks[2].attn.register_forward_hook(get_attention("attn_1"))
+        pretrained.model.blocks[5].attn.register_forward_hook(get_attention("attn_2"))
+        pretrained.model.blocks[8].attn.register_forward_hook(get_attention("attn_3"))
+        pretrained.model.blocks[11].attn.register_forward_hook(get_attention("attn_4"))
+        pretrained.attention = attention
+
+    pretrained.activations = activations
+
+    readout_oper = get_readout_oper(vit_features, features, use_readout, start_index)
+
+    if use_vit_only == True:
+        pretrained.act_postprocess1 = nn.Sequential(
+            readout_oper[0],
+            Transpose(1, 2),
+            nn.Unflatten(2, torch.Size([size[0] // 16, size[1] // 16])),
+            nn.Conv2d(
+                in_channels=vit_features,
+                out_channels=features[0],
+                kernel_size=1,
+                stride=1,
+                padding=0,
+            ),
+            nn.ConvTranspose2d(
+                in_channels=features[0],
+                out_channels=features[0],
+                kernel_size=4,
+                stride=4,
+                padding=0,
+                bias=True,
+                dilation=1,
+                groups=1,
+            ),
+        )
+
+        pretrained.act_postprocess2 = nn.Sequential(
+            readout_oper[1],
+            Transpose(1, 2),
+            nn.Unflatten(2, torch.Size([size[0] // 16, size[1] // 16])),
+            nn.Conv2d(
+                in_channels=vit_features,
+                out_channels=features[1],
+                kernel_size=1,
+                stride=1,
+                padding=0,
+            ),
+            nn.ConvTranspose2d(
+                in_channels=features[1],
+                out_channels=features[1],
+                kernel_size=2,
+                stride=2,
+                padding=0,
+                bias=True,
+                dilation=1,
+                groups=1,
+            ),
+        )
+    else:
+        pretrained.act_postprocess1 = nn.Sequential(
+            nn.Identity(), nn.Identity(), nn.Identity()
+        )
+        pretrained.act_postprocess2 = nn.Sequential(
+            nn.Identity(), nn.Identity(), nn.Identity()
+        )
+
+    pretrained.act_postprocess3 = nn.Sequential(
+        readout_oper[2],
+        Transpose(1, 2),
+        nn.Unflatten(2, torch.Size([size[0] // 16, size[1] // 16])),
+        nn.Conv2d(
+            in_channels=vit_features,
+            out_channels=features[2],
+            kernel_size=1,
+            stride=1,
+            padding=0,
+        ),
+    )
+
+    pretrained.act_postprocess4 = nn.Sequential(
+        readout_oper[3],
+        Transpose(1, 2),
+        nn.Unflatten(2, torch.Size([size[0] // 16, size[1] // 16])),
+        nn.Conv2d(
+            in_channels=vit_features,
+            out_channels=features[3],
+            kernel_size=1,
+            stride=1,
+            padding=0,
+        ),
+        nn.Conv2d(
+            in_channels=features[3],
+            out_channels=features[3],
+            kernel_size=3,
+            stride=2,
+            padding=1,
+        ),
+    )
+
+    pretrained.model.start_index = start_index
+    pretrained.model.patch_size = [16, 16]
+
+    # We inject this function into the VisionTransformer instances so that
+    # we can use it with interpolated position embeddings without modifying the library source.
+    pretrained.model.forward_flex = types.MethodType(forward_flex, pretrained.model)
+
+    # We inject this function into the VisionTransformer instances so that
+    # we can use it with interpolated position embeddings without modifying the library source.
+    pretrained.model._resize_pos_embed = types.MethodType(
+        _resize_pos_embed, pretrained.model
+    )
+
+    return pretrained
+
+
+def _make_pretrained_vitb_rn50_384(
+    pretrained,
+    use_readout="ignore",
+    hooks=None,
+    use_vit_only=False,
+    enable_attention_hooks=False,
+):
+    model = timm.create_model("vit_base_resnet50_384", pretrained=pretrained)
+
+    hooks = [0, 1, 8, 11] if hooks == None else hooks
+    return _make_vit_b_rn50_backbone(
+        model,
+        features=[256, 512, 768, 768],
+        size=[384, 384],
+        hooks=hooks,
+        use_vit_only=use_vit_only,
+        use_readout=use_readout,
+        enable_attention_hooks=enable_attention_hooks,
+    )
+
+
+def _make_pretrained_vitl16_384(
+    pretrained, use_readout="ignore", hooks=None, enable_attention_hooks=False
+):
+    model = timm.create_model("vit_large_patch16_384", pretrained=pretrained)
+
+    hooks = [5, 11, 17, 23] if hooks == None else hooks
+    return _make_vit_b16_backbone(
+        model,
+        features=[256, 512, 1024, 1024],
+        hooks=hooks,
+        vit_features=1024,
+        use_readout=use_readout,
+        enable_attention_hooks=enable_attention_hooks,
+    )
+
+
+def _make_pretrained_vitb16_384(
+    pretrained, use_readout="ignore", hooks=None, enable_attention_hooks=False
+):
+    model = timm.create_model("vit_base_patch16_384", pretrained=pretrained)
+
+    hooks = [2, 5, 8, 11] if hooks == None else hooks
+    return _make_vit_b16_backbone(
+        model,
+        features=[96, 192, 384, 768],
+        hooks=hooks,
+        use_readout=use_readout,
+        enable_attention_hooks=enable_attention_hooks,
+    )
+
+
+def _make_pretrained_deitb16_384(
+    pretrained, use_readout="ignore", hooks=None, enable_attention_hooks=False
+):
+    model = timm.create_model("vit_deit_base_patch16_384", pretrained=pretrained)
+
+    hooks = [2, 5, 8, 11] if hooks == None else hooks
+    return _make_vit_b16_backbone(
+        model,
+        features=[96, 192, 384, 768],
+        hooks=hooks,
+        use_readout=use_readout,
+        enable_attention_hooks=enable_attention_hooks,
+    )
+
+
+def _make_pretrained_deitb16_distil_384(
+    pretrained, use_readout="ignore", hooks=None, enable_attention_hooks=False
+):
+    model = timm.create_model(
+        "vit_deit_base_distilled_patch16_384", pretrained=pretrained
+    )
+
+    hooks = [2, 5, 8, 11] if hooks == None else hooks
+    return _make_vit_b16_backbone(
+        model,
+        features=[96, 192, 384, 768],
+        hooks=hooks,
+        use_readout=use_readout,
+        start_index=2,
+        enable_attention_hooks=enable_attention_hooks,
+    )

DPT/requirements.txt

@@ -0,0 +1,4 @@
+torch==1.8.1
+torchvision==0.9.1
+opencv-python==4.5.2.54
+timm==0.4.5

DPT/run_monodepth.py

@@ -0,0 +1,238 @@
+"""Compute depth maps for images in the input folder.
+"""
+import os
+import glob
+import torch
+import cv2
+import argparse
+
+import util.io
+
+from torchvision.transforms import Compose
+
+from dpt.models import DPTDepthModel
+from dpt.midas_net import MidasNet_large
+from dpt.transforms import Resize, NormalizeImage, PrepareForNet
+
+#from util.misc import visualize_attention
+
+
+def run(input_path, output_path, model_path, model_type="dpt_hybrid", optimize=True):
+    """Run MonoDepthNN to compute depth maps.
+
+    Args:
+        input_path (str): path to input folder
+        output_path (str): path to output folder
+        model_path (str): path to saved model
+    """
+    print("initialize")
+
+    # select device
+    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+    print("device: %s" % device)
+
+    # load network
+    if model_type == "dpt_large":  # DPT-Large
+        net_w = net_h = 384
+        model = DPTDepthModel(
+            path=model_path,
+            backbone="vitl16_384",
+            non_negative=True,
+            enable_attention_hooks=False,
+        )
+        normalization = NormalizeImage(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
+    elif model_type == "dpt_hybrid":  # DPT-Hybrid
+        net_w = net_h = 384
+        model = DPTDepthModel(
+            path=model_path,
+            backbone="vitb_rn50_384",
+            non_negative=True,
+            enable_attention_hooks=False,
+        )
+        normalization = NormalizeImage(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
+    elif model_type == "dpt_hybrid_kitti":
+        net_w = 1216
+        net_h = 352
+
+        model = DPTDepthModel(
+            path=model_path,
+            scale=0.00006016,
+            shift=0.00579,
+            invert=True,
+            backbone="vitb_rn50_384",
+            non_negative=True,
+            enable_attention_hooks=False,
+        )
+
+        normalization = NormalizeImage(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
+    elif model_type == "dpt_hybrid_nyu":
+        net_w = 640
+        net_h = 480
+
+        model = DPTDepthModel(
+            path=model_path,
+            scale=0.000305,
+            shift=0.1378,
+            invert=True,
+            backbone="vitb_rn50_384",
+            non_negative=True,
+            enable_attention_hooks=False,
+        )
+
+        normalization = NormalizeImage(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
+    elif model_type == "midas_v21":  # Convolutional model
+        net_w = net_h = 384
+
+        model = MidasNet_large(model_path, non_negative=True)
+        normalization = NormalizeImage(
+            mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]
+        )
+    else:
+        assert (
+            False
+        ), f"model_type '{model_type}' not implemented, use: --model_type [dpt_large|dpt_hybrid|dpt_hybrid_kitti|dpt_hybrid_nyu|midas_v21]"
+
+    transform = Compose(
+        [
+            Resize(
+                net_w,
+                net_h,
+                resize_target=None,
+                keep_aspect_ratio=True,
+                ensure_multiple_of=32,
+                resize_method="minimal",
+                image_interpolation_method=cv2.INTER_CUBIC,
+            ),
+            normalization,
+            PrepareForNet(),
+        ]
+    )
+
+    model.eval()
+
+    if optimize == True and device == torch.device("cuda"):
+        model = model.to(memory_format=torch.channels_last)
+        model = model.half()
+
+    model.to(device)
+
+    # get input
+    img_names = glob.glob(os.path.join(input_path, "*"))
+    num_images = len(img_names)
+
+    # create output folder
+    os.makedirs(output_path, exist_ok=True)
+
+    print("start processing")
+    for ind, img_name in enumerate(img_names):
+        if os.path.isdir(img_name):
+            continue
+
+        print("  processing {} ({}/{})".format(img_name, ind + 1, num_images))
+        # input
+
+        img = util.io.read_image(img_name)
+
+        if args.kitti_crop is True:
+            height, width, _ = img.shape
+            top = height - 352
+            left = (width - 1216) // 2
+            img = img[top : top + 352, left : left + 1216, :]
+
+        img_input = transform({"image": img})["image"]
+
+        # compute
+        with torch.no_grad():
+            sample = torch.from_numpy(img_input).to(device).unsqueeze(0)
+
+            if optimize == True and device == torch.device("cuda"):
+                sample = sample.to(memory_format=torch.channels_last)
+                sample = sample.half()
+
+            prediction = model.forward(sample)
+            prediction = (
+                torch.nn.functional.interpolate(
+                    prediction.unsqueeze(1),
+                    size=img.shape[:2],
+                    mode="bicubic",
+                    align_corners=False,
+                )
+                .squeeze()
+                .cpu()
+                .numpy()
+            )
+
+            if model_type == "dpt_hybrid_kitti":
+                prediction *= 256
+
+            if model_type == "dpt_hybrid_nyu":
+                prediction *= 1000.0
+
+        filename = os.path.join(
+            output_path, os.path.splitext(os.path.basename(img_name))[0]
+        )
+        util.io.write_depth(filename, prediction, bits=2, absolute_depth=args.absolute_depth)
+
+    print("finished")
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "-i", "--input_path", default="input", help="folder with input images"
+    )
+
+    parser.add_argument(
+        "-o",
+        "--output_path",
+        default="output_monodepth",
+        help="folder for output images",
+    )
+
+    parser.add_argument(
+        "-m", "--model_weights", default=None, help="path to model weights"
+    )
+
+    parser.add_argument(
+        "-t",
+        "--model_type",
+        default="dpt_hybrid",
+        help="model type [dpt_large|dpt_hybrid|midas_v21]",
+    )
+
+    parser.add_argument("--kitti_crop", dest="kitti_crop", action="store_true")
+    parser.add_argument("--absolute_depth", dest="absolute_depth", action="store_true")
+
+    parser.add_argument("--optimize", dest="optimize", action="store_true")
+    parser.add_argument("--no-optimize", dest="optimize", action="store_false")
+
+    parser.set_defaults(optimize=True)
+    parser.set_defaults(kitti_crop=False)
+    parser.set_defaults(absolute_depth=False)
+
+    args = parser.parse_args()
+
+    default_models = {
+        "midas_v21": "weights/midas_v21-f6b98070.pt",
+        "dpt_large": "weights/dpt_large-midas-2f21e586.pt",
+        "dpt_hybrid": "weights/dpt_hybrid-midas-501f0c75.pt",
+        "dpt_hybrid_kitti": "weights/dpt_hybrid_kitti-cb926ef4.pt",
+        "dpt_hybrid_nyu": "weights/dpt_hybrid_nyu-2ce69ec7.pt",
+    }
+
+    if args.model_weights is None:
+        args.model_weights = default_models[args.model_type]
+
+    # set torch options
+    torch.backends.cudnn.enabled = True
+    torch.backends.cudnn.benchmark = True
+
+    # compute depth maps
+    run(
+        args.input_path,
+        args.output_path,
+        args.model_weights,
+        args.model_type,
+        args.optimize,
+    )

DPT/run_segmentation.py

@@ -0,0 +1,163 @@
+"""Compute segmentation maps for images in the input folder.
+"""
+import os
+import glob
+import cv2
+import argparse
+
+import torch
+import torch.nn.functional as F
+
+import util.io
+
+from torchvision.transforms import Compose
+from dpt.models import DPTSegmentationModel
+from dpt.transforms import Resize, NormalizeImage, PrepareForNet
+
+
+def run(input_path, output_path, model_path, model_type="dpt_hybrid", optimize=True):
+    """Run segmentation network
+
+    Args:
+        input_path (str): path to input folder
+        output_path (str): path to output folder
+        model_path (str): path to saved model
+    """
+    print("initialize")
+
+    # select device
+    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+    print("device: %s" % device)
+
+    net_w = net_h = 480
+
+    # load network
+    if model_type == "dpt_large":
+        model = DPTSegmentationModel(
+            150,
+            path=model_path,
+            backbone="vitl16_384",
+        )
+    elif model_type == "dpt_hybrid":
+        model = DPTSegmentationModel(
+            150,
+            path=model_path,
+            backbone="vitb_rn50_384",
+        )
+    else:
+        assert (
+            False
+        ), f"model_type '{model_type}' not implemented, use: --model_type [dpt_large|dpt_hybrid]"
+
+    transform = Compose(
+        [
+            Resize(
+                net_w,
+                net_h,
+                resize_target=None,
+                keep_aspect_ratio=True,
+                ensure_multiple_of=32,
+                resize_method="minimal",
+                image_interpolation_method=cv2.INTER_CUBIC,
+            ),
+            NormalizeImage(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5]),
+            PrepareForNet(),
+        ]
+    )
+
+    model.eval()
+
+    if optimize == True and device == torch.device("cuda"):
+        model = model.to(memory_format=torch.channels_last)
+        model = model.half()
+
+    model.to(device)
+
+    # get input
+    img_names = glob.glob(os.path.join(input_path, "*"))
+    num_images = len(img_names)
+
+    # create output folder
+    os.makedirs(output_path, exist_ok=True)
+
+    print("start processing")
+
+    for ind, img_name in enumerate(img_names):
+
+        print("  processing {} ({}/{})".format(img_name, ind + 1, num_images))
+
+        # input
+        img = util.io.read_image(img_name)
+        img_input = transform({"image": img})["image"]
+
+        # compute
+        with torch.no_grad():
+            sample = torch.from_numpy(img_input).to(device).unsqueeze(0)
+            if optimize == True and device == torch.device("cuda"):
+                sample = sample.to(memory_format=torch.channels_last)
+                sample = sample.half()
+
+            out = model.forward(sample)
+
+            prediction = torch.nn.functional.interpolate(
+                out, size=img.shape[:2], mode="bicubic", align_corners=False
+            )
+            prediction = torch.argmax(prediction, dim=1) + 1
+            prediction = prediction.squeeze().cpu().numpy()
+
+        # output
+        filename = os.path.join(
+            output_path, os.path.splitext(os.path.basename(img_name))[0]
+        )
+        util.io.write_segm_img(filename, img, prediction, alpha=0.5)
+
+    print("finished")
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+
+    parser.add_argument(
+        "-i", "--input_path", default="input", help="folder with input images"
+    )
+
+    parser.add_argument(
+        "-o", "--output_path", default="output_semseg", help="folder for output images"
+    )
+
+    parser.add_argument(
+        "-m",
+        "--model_weights",
+        default=None,
+        help="path to the trained weights of model",
+    )
+
+    # 'vit_large', 'vit_hybrid'
+    parser.add_argument("-t", "--model_type", default="dpt_hybrid", help="model type")
+
+    parser.add_argument("--optimize", dest="optimize", action="store_true")
+    parser.add_argument("--no-optimize", dest="optimize", action="store_false")
+    parser.set_defaults(optimize=True)
+
+    args = parser.parse_args()
+
+    default_models = {
+        "dpt_large": "weights/dpt_large-ade20k-b12dca68.pt",
+        "dpt_hybrid": "weights/dpt_hybrid-ade20k-53898607.pt",
+    }
+
+    if args.model_weights is None:
+        args.model_weights = default_models[args.model_type]
+
+    # set torch options
+    torch.backends.cudnn.enabled = True
+    torch.backends.cudnn.benchmark = True
+
+    # compute segmentation maps
+    run(
+        args.input_path,
+        args.output_path,
+        args.model_weights,
+        args.model_type,
+        args.optimize,
+    )

DPT/setup.py

@@ -0,0 +1,11 @@
+import setuptools
+
+__version__ = '0.0.1dev1'
+
+setuptools.setup(
+    name='dpt',
+    version=__version__,
+    packages=setuptools.find_packages(),
+    # Only put dependencies that's not depends on cuda directly.
+    install_requires=['timm']
+)

DPT/util/io.py

@@ -0,0 +1,219 @@
+"""Utils for monoDepth.
+"""
+import sys
+import re
+import numpy as np
+import cv2
+import torch
+
+from PIL import Image
+
+
+from .pallete import get_mask_pallete
+
+def read_pfm(path):
+    """Read pfm file.
+
+    Args:
+        path (str): path to file
+
+    Returns:
+        tuple: (data, scale)
+    """
+    with open(path, "rb") as file:
+
+        color = None
+        width = None
+        height = None
+        scale = None
+        endian = None
+
+        header = file.readline().rstrip()
+        if header.decode("ascii") == "PF":
+            color = True
+        elif header.decode("ascii") == "Pf":
+            color = False
+        else:
+            raise Exception("Not a PFM file: " + path)
+
+        dim_match = re.match(r"^(\d+)\s(\d+)\s$", file.readline().decode("ascii"))
+        if dim_match:
+            width, height = list(map(int, dim_match.groups()))
+        else:
+            raise Exception("Malformed PFM header.")
+
+        scale = float(file.readline().decode("ascii").rstrip())
+        if scale < 0:
+            # little-endian
+            endian = "<"
+            scale = -scale
+        else:
+            # big-endian
+            endian = ">"
+
+        data = np.fromfile(file, endian + "f")
+        shape = (height, width, 3) if color else (height, width)
+
+        data = np.reshape(data, shape)
+        data = np.flipud(data)
+
+        return data, scale
+
+
+def write_pfm(path, image, scale=1):
+    """Write pfm file.
+
+    Args:
+        path (str): pathto file
+        image (array): data
+        scale (int, optional): Scale. Defaults to 1.
+    """
+
+    with open(path, "wb") as file:
+        color = None
+
+        if image.dtype.name != "float32":
+            raise Exception("Image dtype must be float32.")
+
+        image = np.flipud(image)
+
+        if len(image.shape) == 3 and image.shape[2] == 3:  # color image
+            color = True
+        elif (
+            len(image.shape) == 2 or len(image.shape) == 3 and image.shape[2] == 1
+        ):  # greyscale
+            color = False
+        else:
+            raise Exception("Image must have H x W x 3, H x W x 1 or H x W dimensions.")
+
+        file.write("PF\n" if color else "Pf\n".encode())
+        file.write("%d %d\n".encode() % (image.shape[1], image.shape[0]))
+
+        endian = image.dtype.byteorder
+
+        if endian == "<" or endian == "=" and sys.byteorder == "little":
+            scale = -scale
+
+        file.write("%f\n".encode() % scale)
+
+        image.tofile(file)
+
+
+def read_image(path):
+    """Read image and output RGB image (0-1).
+
+    Args:
+        path (str): path to file
+
+    Returns:
+        array: RGB image (0-1)
+    """
+    img = cv2.imread(path)
+
+    if img.ndim == 2:
+        img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
+
+    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) / 255.0
+
+    return img
+
+
+def resize_image(img):
+    """Resize image and make it fit for network.
+
+    Args:
+        img (array): image
+
+    Returns:
+        tensor: data ready for network
+    """
+    height_orig = img.shape[0]
+    width_orig = img.shape[1]
+
+    if width_orig > height_orig:
+        scale = width_orig / 384
+    else:
+        scale = height_orig / 384
+
+    height = (np.ceil(height_orig / scale / 32) * 32).astype(int)
+    width = (np.ceil(width_orig / scale / 32) * 32).astype(int)
+
+    img_resized = cv2.resize(img, (width, height), interpolation=cv2.INTER_AREA)
+
+    img_resized = (
+        torch.from_numpy(np.transpose(img_resized, (2, 0, 1))).contiguous().float()
+    )
+    img_resized = img_resized.unsqueeze(0)
+
+    return img_resized
+
+
+def resize_depth(depth, width, height):
+    """Resize depth map and bring to CPU (numpy).
+
+    Args:
+        depth (tensor): depth
+        width (int): image width
+        height (int): image height
+
+    Returns:
+        array: processed depth
+    """
+    depth = torch.squeeze(depth[0, :, :, :]).to("cpu")
+
+    depth_resized = cv2.resize(
+        depth.numpy(), (width, height), interpolation=cv2.INTER_CUBIC
+    )
+
+    return depth_resized
+
+
+def write_depth(path, depth, bits=1, absolute_depth=False):
+    """Write depth map to pfm and png file.
+
+    Args:
+        path (str): filepath without extension
+        depth (array): depth
+    """
+    write_pfm(path + ".pfm", depth.astype(np.float32))
+
+    if absolute_depth:
+        out = depth
+    else:
+        depth_min = depth.min()
+        depth_max = depth.max()
+
+        max_val = (2 ** (8 * bits)) - 1
+
+        if depth_max - depth_min > np.finfo("float").eps:
+            out = max_val * (depth - depth_min) / (depth_max - depth_min)
+        else:
+            out = np.zeros(depth.shape, dtype=depth.dtype)
+
+    if bits == 1:
+        cv2.imwrite(path + ".png", out.astype("uint8"), [cv2.IMWRITE_PNG_COMPRESSION, 0])
+    elif bits == 2:
+        cv2.imwrite(path + ".png", out.astype("uint16"), [cv2.IMWRITE_PNG_COMPRESSION, 0])
+
+    return
+
+
+def write_segm_img(path, image, labels, palette="detail", alpha=0.5):
+    """Write depth map to pfm and png file.
+
+    Args:
+        path (str): filepath without extension
+        image (array): input image
+        labels (array): labeling of the image
+    """
+
+    mask = get_mask_pallete(labels, "ade20k")
+
+    img = Image.fromarray(np.uint8(255*image)).convert("RGBA")
+    seg = mask.convert("RGBA")
+
+    out = Image.blend(img, seg, alpha)
+
+    out.save(path + ".png")
+
+    return

DPT/util/misc.py

@@ -0,0 +1,63 @@
+import matplotlib.pyplot as plt
+
+from dpt.vit import get_mean_attention_map
+
+def visualize_attention(input, model, prediction, model_type):
+    input = (input + 1.0)/2.0
+
+    attn1 = model.pretrained.attention["attn_1"]
+    attn2 = model.pretrained.attention["attn_2"]
+    attn3 = model.pretrained.attention["attn_3"]
+    attn4 = model.pretrained.attention["attn_4"]
+
+    plt.subplot(3,4,1), plt.imshow(input.squeeze().permute(1,2,0)), plt.title("Input", fontsize=8), plt.axis("off")
+    plt.subplot(3,4,2), plt.imshow(prediction), plt.set_cmap("inferno"), plt.title("Prediction", fontsize=8), plt.axis("off")
+
+    if model_type == "dpt_hybrid":
+        h = [3,6,9,12]
+    else:
+        h = [6,12,18,24]
+
+    # upper left
+    plt.subplot(345),
+    ax1 = plt.imshow(get_mean_attention_map(attn1, 1, input.shape))
+    plt.ylabel("Upper left corner", fontsize=8)
+    plt.title(f"Layer {h[0]}", fontsize=8)
+    gc = plt.gca()
+    gc.axes.xaxis.set_ticklabels([])
+    gc.axes.yaxis.set_ticklabels([])
+    gc.axes.xaxis.set_ticks([])
+    gc.axes.yaxis.set_ticks([])
+
+
+    plt.subplot(346),
+    plt.imshow(get_mean_attention_map(attn2, 1, input.shape))
+    plt.title(f"Layer {h[1]}", fontsize=8)
+    plt.axis("off"),
+
+    plt.subplot(347),
+    plt.imshow(get_mean_attention_map(attn3, 1, input.shape))
+    plt.title(f"Layer {h[2]}", fontsize=8)
+    plt.axis("off"),
+
+
+    plt.subplot(348),
+    plt.imshow(get_mean_attention_map(attn4, 1, input.shape))
+    plt.title(f"Layer {h[3]}", fontsize=8)
+    plt.axis("off"),
+
+
+    # lower right
+    plt.subplot(3,4,9), plt.imshow(get_mean_attention_map(attn1, -1, input.shape))
+    plt.ylabel("Lower right corner", fontsize=8)
+    gc = plt.gca()
+    gc.axes.xaxis.set_ticklabels([])
+    gc.axes.yaxis.set_ticklabels([])
+    gc.axes.xaxis.set_ticks([])
+    gc.axes.yaxis.set_ticks([])
+
+    plt.subplot(3,4,10), plt.imshow(get_mean_attention_map(attn2, -1, input.shape)), plt.axis("off")
+    plt.subplot(3,4,11), plt.imshow(get_mean_attention_map(attn3, -1, input.shape)), plt.axis("off")
+    plt.subplot(3,4,12), plt.imshow(get_mean_attention_map(attn4, -1, input.shape)), plt.axis("off")
+    plt.tight_layout()
+    plt.show()

DPT/util/pallete.py

@@ -0,0 +1,50 @@
+##+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
+## Created by: Hang Zhang
+## ECE Department, Rutgers University
+## Email: zhang.hang@rutgers.edu
+## Copyright (c) 2017
+##
+## This source code is licensed under the MIT-style license found in the
+## LICENSE file in the root directory of this source tree
+##+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
+
+from PIL import Image
+
+def get_mask_pallete(npimg, dataset='detail'):
+    """Get image color pallete for visualizing masks"""
+    # recovery boundary
+    if dataset == 'pascal_voc':
+        npimg[npimg==21] = 255
+    # put colormap
+    out_img = Image.fromarray(npimg.squeeze().astype('uint8'))
+    if dataset == 'ade20k':
+        out_img.putpalette(adepallete)
+    elif dataset == 'citys':
+        out_img.putpalette(citypallete)
+    elif dataset in ('detail', 'pascal_voc', 'pascal_aug'):
+        out_img.putpalette(vocpallete)
+    return out_img
+
+def _get_voc_pallete(num_cls):
+    n = num_cls
+    pallete = [0]*(n*3)
+    for j in range(0,n):
+            lab = j
+            pallete[j*3+0] = 0
+            pallete[j*3+1] = 0
+            pallete[j*3+2] = 0
+            i = 0
+            while (lab > 0):
+                    pallete[j*3+0] |= (((lab >> 0) & 1) << (7-i))
+                    pallete[j*3+1] |= (((lab >> 1) & 1) << (7-i))
+                    pallete[j*3+2] |= (((lab >> 2) & 1) << (7-i))
+                    i = i + 1
+                    lab >>= 3
+    return pallete
+
+vocpallete = _get_voc_pallete(256)
+
+adepallete = [0,0,0,120,120,120,180,120,120,6,230,230,80,50,50,4,200,3,120,120,80,140,140,140,204,5,255,230,230,230,4,250,7,224,5,255,235,255,7,150,5,61,120,120,70,8,255,51,255,6,82,143,255,140,204,255,4,255,51,7,204,70,3,0,102,200,61,230,250,255,6,51,11,102,255,255,7,71,255,9,224,9,7,230,220,220,220,255,9,92,112,9,255,8,255,214,7,255,224,255,184,6,10,255,71,255,41,10,7,255,255,224,255,8,102,8,255,255,61,6,255,194,7,255,122,8,0,255,20,255,8,41,255,5,153,6,51,255,235,12,255,160,150,20,0,163,255,140,140,140,250,10,15,20,255,0,31,255,0,255,31,0,255,224,0,153,255,0,0,0,255,255,71,0,0,235,255,0,173,255,31,0,255,11,200,200,255,82,0,0,255,245,0,61,255,0,255,112,0,255,133,255,0,0,255,163,0,255,102,0,194,255,0,0,143,255,51,255,0,0,82,255,0,255,41,0,255,173,10,0,255,173,255,0,0,255,153,255,92,0,255,0,255,255,0,245,255,0,102,255,173,0,255,0,20,255,184,184,0,31,255,0,255,61,0,71,255,255,0,204,0,255,194,0,255,82,0,10,255,0,112,255,51,0,255,0,194,255,0,122,255,0,255,163,255,153,0,0,255,10,255,112,0,143,255,0,82,0,255,163,255,0,255,235,0,8,184,170,133,0,255,0,255,92,184,0,255,255,0,31,0,184,255,0,214,255,255,0,112,92,255,0,0,224,255,112,224,255,70,184,160,163,0,255,153,0,255,71,255,0,255,0,163,255,204,0,255,0,143,0,255,235,133,255,0,255,0,235,245,0,255,255,0,122,255,245,0,10,190,212,214,255,0,0,204,255,20,0,255,255,255,0,0,153,255,0,41,255,0,255,204,41,0,255,41,255,0,173,0,255,0,245,255,71,0,255,122,0,255,0,255,184,0,92,255,184,255,0,0,133,255,255,214,0,25,194,194,102,255,0,92,0,255]
+
+citypallete = [
+128,64,128,244,35,232,70,70,70,102,102,156,190,153,153,153,153,153,250,170,30,220,220,0,107,142,35,152,251,152,70,130,180,220,20,60,255,0,0,0,0,142,0,0,70,0,60,100,0,80,100,0,0,230,119,11,32,128,192,0,0,64,128,128,64,128,0,192,128,128,192,128,64,64,0,192,64,0,64,192,0,192,192,0,64,64,128,192,64,128,64,192,128,192,192,128,0,0,64,128,0,64,0,128,64,128,128,64,0,0,192,128,0,192,0,128,192,128,128,192,64,0,64,192,0,64,64,128,64,192,128,64,64,0,192,192,0,192,64,128,192,192,128,192,0,64,64,128,64,64,0,192,64,128,192,64,0,64,192,128,64,192,0,192,192,128,192,192,64,64,64,192,64,64,64,192,64,192,192,64,64,64,192,192,64,192,64,192,192,192,192,192,32,0,0,160,0,0,32,128,0,160,128,0,32,0,128,160,0,128,32,128,128,160,128,128,96,0,0,224,0,0,96,128,0,224,128,0,96,0,128,224,0,128,96,128,128,224,128,128,32,64,0,160,64,0,32,192,0,160,192,0,32,64,128,160,64,128,32,192,128,160,192,128,96,64,0,224,64,0,96,192,0,224,192,0,96,64,128,224,64,128,96,192,128,224,192,128,32,0,64,160,0,64,32,128,64,160,128,64,32,0,192,160,0,192,32,128,192,160,128,192,96,0,64,224,0,64,96,128,64,224,128,64,96,0,192,224,0,192,96,128,192,224,128,192,32,64,64,160,64,64,32,192,64,160,192,64,32,64,192,160,64,192,32,192,192,160,192,192,96,64,64,224,64,64,96,192,64,224,192,64,96,64,192,224,64,192,96,192,192,224,192,192,0,32,0,128,32,0,0,160,0,128,160,0,0,32,128,128,32,128,0,160,128,128,160,128,64,32,0,192,32,0,64,160,0,192,160,0,64,32,128,192,32,128,64,160,128,192,160,128,0,96,0,128,96,0,0,224,0,128,224,0,0,96,128,128,96,128,0,224,128,128,224,128,64,96,0,192,96,0,64,224,0,192,224,0,64,96,128,192,96,128,64,224,128,192,224,128,0,32,64,128,32,64,0,160,64,128,160,64,0,32,192,128,32,192,0,160,192,128,160,192,64,32,64,192,32,64,64,160,64,192,160,64,64,32,192,192,32,192,64,160,192,192,160,192,0,96,64,128,96,64,0,224,64,128,224,64,0,96,192,128,96,192,0,224,192,128,224,192,64,96,64,192,96,64,64,224,64,192,224,64,64,96,192,192,96,192,64,224,192,192,224,192,32,32,0,160,32,0,32,160,0,160,160,0,32,32,128,160,32,128,32,160,128,160,160,128,96,32,0,224,32,0,96,160,0,224,160,0,96,32,128,224,32,128,96,160,128,224,160,128,32,96,0,160,96,0,32,224,0,160,224,0,32,96,128,160,96,128,32,224,128,160,224,128,96,96,0,224,96,0,96,224,0,224,224,0,96,96,128,224,96,128,96,224,128,224,224,128,32,32,64,160,32,64,32,160,64,160,160,64,32,32,192,160,32,192,32,160,192,160,160,192,96,32,64,224,32,64,96,160,64,224,160,64,96,32,192,224,32,192,96,160,192,224,160,192,32,96,64,160,96,64,32,224,64,160,224,64,32,96,192,160,96,192,32,224,192,160,224,192,96,96,64,224,96,64,96,224,64,224,224,64,96,96,192,224,96,192,96,224,192,0,0,0]

demo_gradio.py

@@ -0,0 +1,170 @@
+import gradio as gr
+from diffusers import StableDiffusionXLControlNetInpaintPipeline, ControlNetModel
+from rembg import remove
+from PIL import Image
+import torch
+from ip_adapter import IPAdapterXL
+from ip_adapter.utils import register_cross_attention_hook, get_net_attn_map, attnmaps2images
+from PIL import Image, ImageChops, ImageEnhance
+import numpy as np
+
+import os
+import glob
+import torch
+import cv2
+import argparse
+
+import DPT.util.io
+
+from torchvision.transforms import Compose
+
+from DPT.dpt.models import DPTDepthModel
+from DPT.dpt.midas_net import MidasNet_large
+from DPT.dpt.transforms import Resize, NormalizeImage, PrepareForNet
+
+"""
+Get ZeST Ready
+"""
+base_model_path = "stabilityai/stable-diffusion-xl-base-1.0"
+image_encoder_path = "models/image_encoder"
+ip_ckpt = "sdxl_models/ip-adapter_sdxl_vit-h.bin"
+controlnet_path = "diffusers/controlnet-depth-sdxl-1.0"
+device = "cuda"
+torch.cuda.empty_cache()
+
+# load SDXL pipeline
+controlnet = ControlNetModel.from_pretrained(controlnet_path, variant="fp16", use_safetensors=True, torch_dtype=torch.float16).to(device)
+pipe = StableDiffusionXLControlNetInpaintPipeline.from_pretrained(
+    base_model_path,
+    controlnet=controlnet,
+    use_safetensors=True,
+    torch_dtype=torch.float16,
+    add_watermarker=False,
+).to(device)
+pipe.unet = register_cross_attention_hook(pipe.unet)
+
+ip_model = IPAdapterXL(pipe, image_encoder_path, ip_ckpt, device)
+
+
+"""
+Get Depth Model Ready
+"""
+model_path = "DPT/weights/dpt_hybrid-midas-501f0c75.pt"
+net_w = net_h = 384
+model = DPTDepthModel(
+    path=model_path,
+    backbone="vitb_rn50_384",
+    non_negative=True,
+    enable_attention_hooks=False,
+)
+normalization = NormalizeImage(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
+
+transform = Compose(
+        [
+            Resize(
+                net_w,
+                net_h,
+                resize_target=None,
+                keep_aspect_ratio=True,
+                ensure_multiple_of=32,
+                resize_method="minimal",
+                image_interpolation_method=cv2.INTER_CUBIC,
+            ),
+            normalization,
+            PrepareForNet(),
+        ]
+    )
+
+model.eval()
+
+
+def greet(input_image, material_exemplar):
+    
+    """
+    Compute depth map from input_image
+    """
+    
+    img = np.array(input_image)
+    
+    img_input = transform({"image": img})["image"]
+
+    # compute
+    with torch.no_grad():
+        sample = torch.from_numpy(img_input).unsqueeze(0)
+
+        # if optimize == True and device == torch.device("cuda"):
+        #     sample = sample.to(memory_format=torch.channels_last)
+        #     sample = sample.half()
+
+        prediction = model.forward(sample)
+        prediction = (
+            torch.nn.functional.interpolate(
+                prediction.unsqueeze(1),
+                size=img.shape[:2],
+                mode="bicubic",
+                align_corners=False,
+            )
+            .squeeze()
+            .cpu()
+            .numpy()
+        )
+        
+    depth_min = prediction.min()
+    depth_max = prediction.max()
+    bits = 2
+    max_val = (2 ** (8 * bits)) - 1
+
+    if depth_max - depth_min > np.finfo("float").eps:
+        out = max_val * (prediction - depth_min) / (depth_max - depth_min)
+    else:
+        out = np.zeros(prediction.shape, dtype=depth.dtype)
+    
+    out = (out / 256).astype('uint8')
+    depth_map = Image.fromarray(out).resize((1024, 1024))
+    
+    
+    """
+    Process foreground decolored image
+    """
+    rm_bg = remove(input_image)
+    target_mask = rm_bg.convert("RGB").point(lambda x: 0 if x < 1 else 255).convert('L').convert('RGB')
+    mask_target_img = ImageChops.lighter(input_image, target_mask)
+    invert_target_mask = ImageChops.invert(target_mask)
+    gray_target_image = input_image.convert('L').convert('RGB')
+    gray_target_image = ImageEnhance.Brightness(gray_target_image)
+    factor = 1.0  # Try adjusting this to get the desired brightness
+    gray_target_image = gray_target_image.enhance(factor)
+    grayscale_img = ImageChops.darker(gray_target_image, target_mask)
+    img_black_mask = ImageChops.darker(input_image, invert_target_mask)
+    grayscale_init_img = ImageChops.lighter(img_black_mask, grayscale_img)
+    init_img = grayscale_init_img
+    
+    """
+    Process material exemplar and resize all images
+    """
+    ip_image = material_exemplar.resize((1024, 1024))
+    init_img = init_img.resize((1024,1024))
+    mask = target_mask.resize((1024, 1024))
+    
+    
+    num_samples = 1
+    images = ip_model.generate(pil_image=ip_image, image=init_img, control_image=depth_map, mask_image=mask, controlnet_conditioning_scale=0.9, num_samples=num_samples, num_inference_steps=30, seed=42)
+    
+    return images[0]
+
+
+
+input_image = gr.Image(type="pil")
+input_image2 = gr.Image(type="pil")
+
+demo = gr.Interface(
+    fn=greet,
+    inputs=[input_image, input_image2],
+    title="ZeST: Zero-Shot Material Transfer from a Single Image",
+    description="Upload two images -- input image and material exemplar. ZeST extracts the material from the exemplar and cast it onto the input image following the original lighting cues.",
+    outputs=["image"],
+    allow_flagging='never'
+
+)
+
+demo.launch()

ip_adapter/__init__.py

@@ -0,0 +1,9 @@
+from .ip_adapter import IPAdapter, IPAdapterPlus, IPAdapterPlusXL, IPAdapterXL, IPAdapterFull
+
+__all__ = [
+    "IPAdapter",
+    "IPAdapterPlus",
+    "IPAdapterPlusXL",
+    "IPAdapterXL",
+    "IPAdapterFull",
+]

ip_adapter/attention_processor.py

@@ -0,0 +1,568 @@
+# modified from https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+
+class AttnProcessor(nn.Module):
+    r"""
+    Default processor for performing attention-related computations.
+    """
+
+    def __init__(
+        self,
+        hidden_size=None,
+        cross_attention_dim=None,
+    ):
+        super().__init__()
+
+    def __call__(
+        self,
+        attn,
+        hidden_states,
+        encoder_hidden_states=None,
+        attention_mask=None,
+        temb=None,
+        *args,
+        **kwargs,
+    ):
+        residual = hidden_states
+
+        if attn.spatial_norm is not None:
+            hidden_states = attn.spatial_norm(hidden_states, temb)
+
+        input_ndim = hidden_states.ndim
+
+        if input_ndim == 4:
+            batch_size, channel, height, width = hidden_states.shape
+            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)
+
+        batch_size, sequence_length, _ = (
+            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
+        )
+        attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
+
+        if attn.group_norm is not None:
+            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)
+
+        query = attn.to_q(hidden_states)
+
+        if encoder_hidden_states is None:
+            encoder_hidden_states = hidden_states
+        elif attn.norm_cross:
+            encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)
+
+        key = attn.to_k(encoder_hidden_states)
+        value = attn.to_v(encoder_hidden_states)
+
+        query = attn.head_to_batch_dim(query)
+        key = attn.head_to_batch_dim(key)
+        value = attn.head_to_batch_dim(value)
+
+        attention_probs = attn.get_attention_scores(query, key, attention_mask)
+        hidden_states = torch.bmm(attention_probs, value)
+        hidden_states = attn.batch_to_head_dim(hidden_states)
+
+        # linear proj
+        hidden_states = attn.to_out[0](hidden_states)
+        # dropout
+        hidden_states = attn.to_out[1](hidden_states)
+
+        if input_ndim == 4:
+            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)
+
+        if attn.residual_connection:
+            hidden_states = hidden_states + residual
+
+        hidden_states = hidden_states / attn.rescale_output_factor
+
+        return hidden_states
+
+
+class IPAttnProcessor(nn.Module):
+    r"""
+    Attention processor for IP-Adapater.
+    Args:
+        hidden_size (`int`):
+            The hidden size of the attention layer.
+        cross_attention_dim (`int`):
+            The number of channels in the `encoder_hidden_states`.
+        scale (`float`, defaults to 1.0):
+            the weight scale of image prompt.
+        num_tokens (`int`, defaults to 4 when do ip_adapter_plus it should be 16):
+            The context length of the image features.
+    """
+
+    def __init__(self, hidden_size, cross_attention_dim=None, scale=1.0, num_tokens=4):
+        super().__init__()
+
+        self.hidden_size = hidden_size
+        self.cross_attention_dim = cross_attention_dim
+        self.scale = scale
+        self.num_tokens = num_tokens
+
+        self.to_k_ip = nn.Linear(cross_attention_dim or hidden_size, hidden_size, bias=False)
+        self.to_v_ip = nn.Linear(cross_attention_dim or hidden_size, hidden_size, bias=False)
+
+    def __call__(
+        self,
+        attn,
+        hidden_states,
+        encoder_hidden_states=None,
+        attention_mask=None,
+        temb=None,
+        *args,
+        **kwargs,
+    ):
+        residual = hidden_states
+
+        if attn.spatial_norm is not None:
+            hidden_states = attn.spatial_norm(hidden_states, temb)
+
+        input_ndim = hidden_states.ndim
+
+        if input_ndim == 4:
+            batch_size, channel, height, width = hidden_states.shape
+            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)
+
+        batch_size, sequence_length, _ = (
+            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
+        )
+        attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
+
+        if attn.group_norm is not None:
+            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)
+
+        query = attn.to_q(hidden_states)
+
+        if encoder_hidden_states is None:
+            encoder_hidden_states = hidden_states
+        else:
+            # get encoder_hidden_states, ip_hidden_states
+            end_pos = encoder_hidden_states.shape[1] - self.num_tokens
+            encoder_hidden_states, ip_hidden_states = (
+                encoder_hidden_states[:, :end_pos, :],
+                encoder_hidden_states[:, end_pos:, :],
+            )
+            if attn.norm_cross:
+                encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)
+
+        key = attn.to_k(encoder_hidden_states)
+        value = attn.to_v(encoder_hidden_states)
+
+        query = attn.head_to_batch_dim(query)
+        key = attn.head_to_batch_dim(key)
+        value = attn.head_to_batch_dim(value)
+
+        attention_probs = attn.get_attention_scores(query, key, attention_mask)
+        hidden_states = torch.bmm(attention_probs, value)
+        hidden_states = attn.batch_to_head_dim(hidden_states)
+
+        # for ip-adapter
+        ip_key = self.to_k_ip(ip_hidden_states)
+        ip_value = self.to_v_ip(ip_hidden_states)
+
+        ip_key = attn.head_to_batch_dim(ip_key)
+        ip_value = attn.head_to_batch_dim(ip_value)
+
+        ip_attention_probs = attn.get_attention_scores(query, ip_key, None)
+        self.attn_map = ip_attention_probs
+        ip_hidden_states = torch.bmm(ip_attention_probs, ip_value)
+        ip_hidden_states = attn.batch_to_head_dim(ip_hidden_states)
+
+        hidden_states = hidden_states + self.scale * ip_hidden_states
+
+        # linear proj
+        hidden_states = attn.to_out[0](hidden_states)
+        # dropout
+        hidden_states = attn.to_out[1](hidden_states)
+
+        if input_ndim == 4:
+            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)
+
+        if attn.residual_connection:
+            hidden_states = hidden_states + residual
+
+        hidden_states = hidden_states / attn.rescale_output_factor
+
+        return hidden_states
+
+
+class AttnProcessor2_0(torch.nn.Module):
+    r"""
+    Processor for implementing scaled dot-product attention (enabled by default if you're using PyTorch 2.0).
+    """
+
+    def __init__(
+        self,
+        hidden_size=None,
+        cross_attention_dim=None,
+    ):
+        super().__init__()
+        if not hasattr(F, "scaled_dot_product_attention"):
+            raise ImportError("AttnProcessor2_0 requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.")
+
+    def __call__(
+        self,
+        attn,
+        hidden_states,
+        encoder_hidden_states=None,
+        attention_mask=None,
+        temb=None,
+        *args,
+        **kwargs,
+    ):
+        residual = hidden_states
+
+        if attn.spatial_norm is not None:
+            hidden_states = attn.spatial_norm(hidden_states, temb)
+
+        input_ndim = hidden_states.ndim
+
+        if input_ndim == 4:
+            batch_size, channel, height, width = hidden_states.shape
+            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)
+
+        batch_size, sequence_length, _ = (
+            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
+        )
+
+        if attention_mask is not None:
+            attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
+            # scaled_dot_product_attention expects attention_mask shape to be
+            # (batch, heads, source_length, target_length)
+            attention_mask = attention_mask.view(batch_size, attn.heads, -1, attention_mask.shape[-1])
+
+        if attn.group_norm is not None:
+            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)
+
+        query = attn.to_q(hidden_states)
+
+        if encoder_hidden_states is None:
+            encoder_hidden_states = hidden_states
+        elif attn.norm_cross:
+            encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)
+
+        key = attn.to_k(encoder_hidden_states)
+        value = attn.to_v(encoder_hidden_states)
+
+        inner_dim = key.shape[-1]
+        head_dim = inner_dim // attn.heads
+
+        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+        # the output of sdp = (batch, num_heads, seq_len, head_dim)
+        # TODO: add support for attn.scale when we move to Torch 2.1
+        hidden_states = F.scaled_dot_product_attention(
+            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
+        )
+
+        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
+        hidden_states = hidden_states.to(query.dtype)
+
+        # linear proj
+        hidden_states = attn.to_out[0](hidden_states)
+        # dropout
+        hidden_states = attn.to_out[1](hidden_states)
+
+        if input_ndim == 4:
+            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)
+
+        if attn.residual_connection:
+            hidden_states = hidden_states + residual
+
+        hidden_states = hidden_states / attn.rescale_output_factor
+
+        return hidden_states
+
+
+class IPAttnProcessor2_0(torch.nn.Module):
+    r"""
+    Attention processor for IP-Adapater for PyTorch 2.0.
+    Args:
+        hidden_size (`int`):
+            The hidden size of the attention layer.
+        cross_attention_dim (`int`):
+            The number of channels in the `encoder_hidden_states`.
+        scale (`float`, defaults to 1.0):
+            the weight scale of image prompt.
+        num_tokens (`int`, defaults to 4 when do ip_adapter_plus it should be 16):
+            The context length of the image features.
+    """
+
+    def __init__(self, hidden_size, cross_attention_dim=None, scale=1.0, num_tokens=4):
+        super().__init__()
+
+        if not hasattr(F, "scaled_dot_product_attention"):
+            raise ImportError("AttnProcessor2_0 requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.")
+
+        self.hidden_size = hidden_size
+        self.cross_attention_dim = cross_attention_dim
+        self.scale = scale
+        self.num_tokens = num_tokens
+
+        self.to_k_ip = nn.Linear(cross_attention_dim or hidden_size, hidden_size, bias=False)
+        self.to_v_ip = nn.Linear(cross_attention_dim or hidden_size, hidden_size, bias=False)
+
+    def __call__(
+        self,
+        attn,
+        hidden_states,
+        encoder_hidden_states=None,
+        attention_mask=None,
+        temb=None,
+        *args,
+        **kwargs,
+    ):
+        residual = hidden_states
+
+        if attn.spatial_norm is not None:
+            hidden_states = attn.spatial_norm(hidden_states, temb)
+
+        input_ndim = hidden_states.ndim
+
+        if input_ndim == 4:
+            batch_size, channel, height, width = hidden_states.shape
+            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)
+
+        batch_size, sequence_length, _ = (
+            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
+        )
+
+        if attention_mask is not None:
+            attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
+            # scaled_dot_product_attention expects attention_mask shape to be
+            # (batch, heads, source_length, target_length)
+            attention_mask = attention_mask.view(batch_size, attn.heads, -1, attention_mask.shape[-1])
+
+        if attn.group_norm is not None:
+            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)
+
+        query = attn.to_q(hidden_states)
+
+        if encoder_hidden_states is None:
+            encoder_hidden_states = hidden_states
+        else:
+            # get encoder_hidden_states, ip_hidden_states
+            end_pos = encoder_hidden_states.shape[1] - self.num_tokens
+            encoder_hidden_states, ip_hidden_states = (
+                encoder_hidden_states[:, :end_pos, :],
+                encoder_hidden_states[:, end_pos:, :],
+            )
+            if attn.norm_cross:
+                encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)
+
+        key = attn.to_k(encoder_hidden_states)
+        value = attn.to_v(encoder_hidden_states)
+
+        inner_dim = key.shape[-1]
+        head_dim = inner_dim // attn.heads
+
+        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+        # the output of sdp = (batch, num_heads, seq_len, head_dim)
+        # TODO: add support for attn.scale when we move to Torch 2.1
+        hidden_states = F.scaled_dot_product_attention(
+            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
+        )
+
+        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
+        hidden_states = hidden_states.to(query.dtype)
+
+        # for ip-adapter
+        ip_key = self.to_k_ip(ip_hidden_states)
+        ip_value = self.to_v_ip(ip_hidden_states)
+
+        ip_key = ip_key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+        ip_value = ip_value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+        # the output of sdp = (batch, num_heads, seq_len, head_dim)
+        # TODO: add support for attn.scale when we move to Torch 2.1
+        ip_hidden_states = F.scaled_dot_product_attention(
+            query, ip_key, ip_value, attn_mask=None, dropout_p=0.0, is_causal=False
+        )
+        with torch.no_grad():
+            self.attn_map = query @ ip_key.transpose(-2, -1).softmax(dim=-1)
+            #print(self.attn_map.shape)
+
+        ip_hidden_states = ip_hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
+        ip_hidden_states = ip_hidden_states.to(query.dtype)
+
+        hidden_states = hidden_states + self.scale * ip_hidden_states
+
+        # linear proj
+        hidden_states = attn.to_out[0](hidden_states)
+        # dropout
+        hidden_states = attn.to_out[1](hidden_states)
+
+        if input_ndim == 4:
+            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)
+
+        if attn.residual_connection:
+            hidden_states = hidden_states + residual
+
+        hidden_states = hidden_states / attn.rescale_output_factor
+
+        return hidden_states
+
+
+## for controlnet
+class CNAttnProcessor:
+    r"""
+    Default processor for performing attention-related computations.
+    """
+
+    def __init__(self, num_tokens=4):
+        self.num_tokens = num_tokens
+
+    def __call__(self, attn, hidden_states, encoder_hidden_states=None, attention_mask=None, temb=None, *args, **kwargs,):
+        residual = hidden_states
+
+        if attn.spatial_norm is not None:
+            hidden_states = attn.spatial_norm(hidden_states, temb)
+
+        input_ndim = hidden_states.ndim
+
+        if input_ndim == 4:
+            batch_size, channel, height, width = hidden_states.shape
+            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)
+
+        batch_size, sequence_length, _ = (
+            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
+        )
+        attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
+
+        if attn.group_norm is not None:
+            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)
+
+        query = attn.to_q(hidden_states)
+
+        if encoder_hidden_states is None:
+            encoder_hidden_states = hidden_states
+        else:
+            end_pos = encoder_hidden_states.shape[1] - self.num_tokens
+            encoder_hidden_states = encoder_hidden_states[:, :end_pos]  # only use text
+            if attn.norm_cross:
+                encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)
+
+        key = attn.to_k(encoder_hidden_states)
+        value = attn.to_v(encoder_hidden_states)
+
+        query = attn.head_to_batch_dim(query)
+        key = attn.head_to_batch_dim(key)
+        value = attn.head_to_batch_dim(value)
+
+        attention_probs = attn.get_attention_scores(query, key, attention_mask)
+        hidden_states = torch.bmm(attention_probs, value)
+        hidden_states = attn.batch_to_head_dim(hidden_states)
+
+        # linear proj
+        hidden_states = attn.to_out[0](hidden_states)
+        # dropout
+        hidden_states = attn.to_out[1](hidden_states)
+
+        if input_ndim == 4:
+            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)
+
+        if attn.residual_connection:
+            hidden_states = hidden_states + residual
+
+        hidden_states = hidden_states / attn.rescale_output_factor
+
+        return hidden_states
+
+
+class CNAttnProcessor2_0:
+    r"""
+    Processor for implementing scaled dot-product attention (enabled by default if you're using PyTorch 2.0).
+    """
+
+    def __init__(self, num_tokens=4):
+        if not hasattr(F, "scaled_dot_product_attention"):
+            raise ImportError("AttnProcessor2_0 requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.")
+        self.num_tokens = num_tokens
+
+    def __call__(
+        self,
+        attn,
+        hidden_states,
+        encoder_hidden_states=None,
+        attention_mask=None,
+        temb=None,
+        *args,
+        **kwargs,
+    ):
+        residual = hidden_states
+
+        if attn.spatial_norm is not None:
+            hidden_states = attn.spatial_norm(hidden_states, temb)
+
+        input_ndim = hidden_states.ndim
+
+        if input_ndim == 4:
+            batch_size, channel, height, width = hidden_states.shape
+            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)
+
+        batch_size, sequence_length, _ = (
+            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
+        )
+
+        if attention_mask is not None:
+            attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
+            # scaled_dot_product_attention expects attention_mask shape to be
+            # (batch, heads, source_length, target_length)
+            attention_mask = attention_mask.view(batch_size, attn.heads, -1, attention_mask.shape[-1])
+
+        if attn.group_norm is not None:
+            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)
+
+        query = attn.to_q(hidden_states)
+
+        if encoder_hidden_states is None:
+            encoder_hidden_states = hidden_states
+        else:
+            end_pos = encoder_hidden_states.shape[1] - self.num_tokens
+            encoder_hidden_states = encoder_hidden_states[:, :end_pos]  # only use text
+            if attn.norm_cross:
+                encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)
+
+        key = attn.to_k(encoder_hidden_states)
+        value = attn.to_v(encoder_hidden_states)
+
+        inner_dim = key.shape[-1]
+        head_dim = inner_dim // attn.heads
+
+        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+        # the output of sdp = (batch, num_heads, seq_len, head_dim)
+        # TODO: add support for attn.scale when we move to Torch 2.1
+        hidden_states = F.scaled_dot_product_attention(
+            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
+        )
+
+        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
+        hidden_states = hidden_states.to(query.dtype)
+
+        # linear proj
+        hidden_states = attn.to_out[0](hidden_states)
+        # dropout
+        hidden_states = attn.to_out[1](hidden_states)
+
+        if input_ndim == 4:
+            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)
+
+        if attn.residual_connection:
+            hidden_states = hidden_states + residual
+
+        hidden_states = hidden_states / attn.rescale_output_factor
+
+        return hidden_states

ip_adapter/attention_processor_faceid.py

@@ -0,0 +1,433 @@
+# modified from https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from diffusers.models.lora import LoRALinearLayer
+
+
+class LoRAAttnProcessor(nn.Module):
+    r"""
+    Default processor for performing attention-related computations.
+    """
+
+    def __init__(
+        self,
+        hidden_size=None,
+        cross_attention_dim=None,
+        rank=4,
+        network_alpha=None,
+        lora_scale=1.0,
+    ):
+        super().__init__()
+
+        self.rank = rank
+        self.lora_scale = lora_scale
+        
+        self.to_q_lora = LoRALinearLayer(hidden_size, hidden_size, rank, network_alpha)
+        self.to_k_lora = LoRALinearLayer(cross_attention_dim or hidden_size, hidden_size, rank, network_alpha)
+        self.to_v_lora = LoRALinearLayer(cross_attention_dim or hidden_size, hidden_size, rank, network_alpha)
+        self.to_out_lora = LoRALinearLayer(hidden_size, hidden_size, rank, network_alpha)
+
+    def __call__(
+        self,
+        attn,
+        hidden_states,
+        encoder_hidden_states=None,
+        attention_mask=None,
+        temb=None,
+        *args,
+        **kwargs,
+    ):
+        residual = hidden_states
+
+        if attn.spatial_norm is not None:
+            hidden_states = attn.spatial_norm(hidden_states, temb)
+
+        input_ndim = hidden_states.ndim
+
+        if input_ndim == 4:
+            batch_size, channel, height, width = hidden_states.shape
+            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)
+
+        batch_size, sequence_length, _ = (
+            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
+        )
+        attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
+
+        if attn.group_norm is not None:
+            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)
+
+        query = attn.to_q(hidden_states) + self.lora_scale * self.to_q_lora(hidden_states)
+
+        if encoder_hidden_states is None:
+            encoder_hidden_states = hidden_states
+        elif attn.norm_cross:
+            encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)
+
+        key = attn.to_k(encoder_hidden_states) + self.lora_scale * self.to_k_lora(encoder_hidden_states)
+        value = attn.to_v(encoder_hidden_states) + self.lora_scale * self.to_v_lora(encoder_hidden_states)
+
+        query = attn.head_to_batch_dim(query)
+        key = attn.head_to_batch_dim(key)
+        value = attn.head_to_batch_dim(value)
+
+        attention_probs = attn.get_attention_scores(query, key, attention_mask)
+        hidden_states = torch.bmm(attention_probs, value)
+        hidden_states = attn.batch_to_head_dim(hidden_states)
+
+        # linear proj
+        hidden_states = attn.to_out[0](hidden_states) + self.lora_scale * self.to_out_lora(hidden_states)
+        # dropout
+        hidden_states = attn.to_out[1](hidden_states)
+
+        if input_ndim == 4:
+            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)
+
+        if attn.residual_connection:
+            hidden_states = hidden_states + residual
+
+        hidden_states = hidden_states / attn.rescale_output_factor
+
+        return hidden_states
+
+
+class LoRAIPAttnProcessor(nn.Module):
+    r"""
+    Attention processor for IP-Adapater.
+    Args:
+        hidden_size (`int`):
+            The hidden size of the attention layer.
+        cross_attention_dim (`int`):
+            The number of channels in the `encoder_hidden_states`.
+        scale (`float`, defaults to 1.0):
+            the weight scale of image prompt.
+        num_tokens (`int`, defaults to 4 when do ip_adapter_plus it should be 16):
+            The context length of the image features.
+    """
+
+    def __init__(self, hidden_size, cross_attention_dim=None, rank=4, network_alpha=None, lora_scale=1.0, scale=1.0, num_tokens=4):
+        super().__init__()
+
+        self.rank = rank
+        self.lora_scale = lora_scale
+        
+        self.to_q_lora = LoRALinearLayer(hidden_size, hidden_size, rank, network_alpha)
+        self.to_k_lora = LoRALinearLayer(cross_attention_dim or hidden_size, hidden_size, rank, network_alpha)
+        self.to_v_lora = LoRALinearLayer(cross_attention_dim or hidden_size, hidden_size, rank, network_alpha)
+        self.to_out_lora = LoRALinearLayer(hidden_size, hidden_size, rank, network_alpha)
+
+        self.hidden_size = hidden_size
+        self.cross_attention_dim = cross_attention_dim
+        self.scale = scale
+        self.num_tokens = num_tokens
+
+        self.to_k_ip = nn.Linear(cross_attention_dim or hidden_size, hidden_size, bias=False)
+        self.to_v_ip = nn.Linear(cross_attention_dim or hidden_size, hidden_size, bias=False)
+
+    def __call__(
+        self,
+        attn,
+        hidden_states,
+        encoder_hidden_states=None,
+        attention_mask=None,
+        temb=None,
+        *args,
+        **kwargs,
+    ):
+        residual = hidden_states
+
+        if attn.spatial_norm is not None:
+            hidden_states = attn.spatial_norm(hidden_states, temb)
+
+        input_ndim = hidden_states.ndim
+
+        if input_ndim == 4:
+            batch_size, channel, height, width = hidden_states.shape
+            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)
+
+        batch_size, sequence_length, _ = (
+            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
+        )
+        attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
+
+        if attn.group_norm is not None:
+            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)
+
+        query = attn.to_q(hidden_states) + self.lora_scale * self.to_q_lora(hidden_states)
+
+        if encoder_hidden_states is None:
+            encoder_hidden_states = hidden_states
+        else:
+            # get encoder_hidden_states, ip_hidden_states
+            end_pos = encoder_hidden_states.shape[1] - self.num_tokens
+            encoder_hidden_states, ip_hidden_states = (
+                encoder_hidden_states[:, :end_pos, :],
+                encoder_hidden_states[:, end_pos:, :],
+            )
+            if attn.norm_cross:
+                encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)
+
+        key = attn.to_k(encoder_hidden_states) + self.lora_scale * self.to_k_lora(encoder_hidden_states)
+        value = attn.to_v(encoder_hidden_states) + self.lora_scale * self.to_v_lora(encoder_hidden_states)
+
+        query = attn.head_to_batch_dim(query)
+        key = attn.head_to_batch_dim(key)
+        value = attn.head_to_batch_dim(value)
+
+        attention_probs = attn.get_attention_scores(query, key, attention_mask)
+        hidden_states = torch.bmm(attention_probs, value)
+        hidden_states = attn.batch_to_head_dim(hidden_states)
+
+        # for ip-adapter
+        ip_key = self.to_k_ip(ip_hidden_states)
+        ip_value = self.to_v_ip(ip_hidden_states)
+
+        ip_key = attn.head_to_batch_dim(ip_key)
+        ip_value = attn.head_to_batch_dim(ip_value)
+
+        ip_attention_probs = attn.get_attention_scores(query, ip_key, None)
+        self.attn_map = ip_attention_probs
+        ip_hidden_states = torch.bmm(ip_attention_probs, ip_value)
+        ip_hidden_states = attn.batch_to_head_dim(ip_hidden_states)
+
+        hidden_states = hidden_states + self.scale * ip_hidden_states
+
+        # linear proj
+        hidden_states = attn.to_out[0](hidden_states) + self.lora_scale * self.to_out_lora(hidden_states)
+        # dropout
+        hidden_states = attn.to_out[1](hidden_states)
+
+        if input_ndim == 4:
+            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)
+
+        if attn.residual_connection:
+            hidden_states = hidden_states + residual
+
+        hidden_states = hidden_states / attn.rescale_output_factor
+
+        return hidden_states
+
+
+class LoRAAttnProcessor2_0(nn.Module):
+    
+    r"""
+    Default processor for performing attention-related computations.
+    """
+    
+    def __init__(
+        self,
+        hidden_size=None,
+        cross_attention_dim=None,
+        rank=4,
+        network_alpha=None,
+        lora_scale=1.0,
+    ):
+        super().__init__()
+        
+        self.rank = rank
+        self.lora_scale = lora_scale
+        
+        self.to_q_lora = LoRALinearLayer(hidden_size, hidden_size, rank, network_alpha)
+        self.to_k_lora = LoRALinearLayer(cross_attention_dim or hidden_size, hidden_size, rank, network_alpha)
+        self.to_v_lora = LoRALinearLayer(cross_attention_dim or hidden_size, hidden_size, rank, network_alpha)
+        self.to_out_lora = LoRALinearLayer(hidden_size, hidden_size, rank, network_alpha)
+
+    def __call__(
+        self,
+        attn,
+        hidden_states,
+        encoder_hidden_states=None,
+        attention_mask=None,
+        temb=None,
+        *args,
+        **kwargs,
+    ):
+        residual = hidden_states
+
+        if attn.spatial_norm is not None:
+            hidden_states = attn.spatial_norm(hidden_states, temb)
+
+        input_ndim = hidden_states.ndim
+
+        if input_ndim == 4:
+            batch_size, channel, height, width = hidden_states.shape
+            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)
+
+        batch_size, sequence_length, _ = (
+            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
+        )
+        attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
+
+        if attn.group_norm is not None:
+            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)
+
+        query = attn.to_q(hidden_states) + self.lora_scale * self.to_q_lora(hidden_states)
+
+        if encoder_hidden_states is None:
+            encoder_hidden_states = hidden_states
+        elif attn.norm_cross:
+            encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)
+
+        key = attn.to_k(encoder_hidden_states) + self.lora_scale * self.to_k_lora(encoder_hidden_states)
+        value = attn.to_v(encoder_hidden_states) + self.lora_scale * self.to_v_lora(encoder_hidden_states)
+
+        inner_dim = key.shape[-1]
+        head_dim = inner_dim // attn.heads
+
+        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+        # the output of sdp = (batch, num_heads, seq_len, head_dim)
+        # TODO: add support for attn.scale when we move to Torch 2.1
+        hidden_states = F.scaled_dot_product_attention(
+            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
+        )
+
+        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
+        hidden_states = hidden_states.to(query.dtype)
+
+        # linear proj
+        hidden_states = attn.to_out[0](hidden_states) + self.lora_scale * self.to_out_lora(hidden_states)
+        # dropout
+        hidden_states = attn.to_out[1](hidden_states)
+
+        if input_ndim == 4:
+            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)
+
+        if attn.residual_connection:
+            hidden_states = hidden_states + residual
+
+        hidden_states = hidden_states / attn.rescale_output_factor
+
+        return hidden_states
+
+
+class LoRAIPAttnProcessor2_0(nn.Module):
+    r"""
+    Processor for implementing the LoRA attention mechanism.
+
+    Args:
+        hidden_size (`int`, *optional*):
+            The hidden size of the attention layer.
+        cross_attention_dim (`int`, *optional*):
+            The number of channels in the `encoder_hidden_states`.
+        rank (`int`, defaults to 4):
+            The dimension of the LoRA update matrices.
+        network_alpha (`int`, *optional*):
+            Equivalent to `alpha` but it's usage is specific to Kohya (A1111) style LoRAs.
+    """
+
+    def __init__(self, hidden_size, cross_attention_dim=None, rank=4, network_alpha=None, lora_scale=1.0, scale=1.0, num_tokens=4):
+        super().__init__()
+        
+        self.rank = rank
+        self.lora_scale = lora_scale
+        self.num_tokens = num_tokens
+        
+        self.to_q_lora = LoRALinearLayer(hidden_size, hidden_size, rank, network_alpha)
+        self.to_k_lora = LoRALinearLayer(cross_attention_dim or hidden_size, hidden_size, rank, network_alpha)
+        self.to_v_lora = LoRALinearLayer(cross_attention_dim or hidden_size, hidden_size, rank, network_alpha)
+        self.to_out_lora = LoRALinearLayer(hidden_size, hidden_size, rank, network_alpha)
+        
+        
+        self.hidden_size = hidden_size
+        self.cross_attention_dim = cross_attention_dim
+        self.scale = scale
+
+        self.to_k_ip = nn.Linear(cross_attention_dim or hidden_size, hidden_size, bias=False)
+        self.to_v_ip = nn.Linear(cross_attention_dim or hidden_size, hidden_size, bias=False)
+
+    def __call__(
+        self, attn, hidden_states, encoder_hidden_states=None, attention_mask=None, scale=1.0, temb=None, *args, **kwargs,
+    ):
+        residual = hidden_states
+
+        if attn.spatial_norm is not None:
+            hidden_states = attn.spatial_norm(hidden_states, temb)
+
+        input_ndim = hidden_states.ndim
+
+        if input_ndim == 4:
+            batch_size, channel, height, width = hidden_states.shape
+            hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)
+
+        batch_size, sequence_length, _ = (
+            hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
+        )
+        attention_mask = attn.prepare_attention_mask(attention_mask, sequence_length, batch_size)
+
+        if attn.group_norm is not None:
+            hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)
+
+        query = attn.to_q(hidden_states) + self.lora_scale * self.to_q_lora(hidden_states)
+        #query = attn.head_to_batch_dim(query)
+
+        if encoder_hidden_states is None:
+            encoder_hidden_states = hidden_states
+        else:
+            # get encoder_hidden_states, ip_hidden_states
+            end_pos = encoder_hidden_states.shape[1] - self.num_tokens
+            encoder_hidden_states, ip_hidden_states = (
+                encoder_hidden_states[:, :end_pos, :],
+                encoder_hidden_states[:, end_pos:, :],
+            )
+            if attn.norm_cross:
+                encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)
+
+        # for text
+        key = attn.to_k(encoder_hidden_states) + self.lora_scale * self.to_k_lora(encoder_hidden_states)
+        value = attn.to_v(encoder_hidden_states) + self.lora_scale * self.to_v_lora(encoder_hidden_states)
+
+        inner_dim = key.shape[-1]
+        head_dim = inner_dim // attn.heads
+
+        query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+        key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+        value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+        # the output of sdp = (batch, num_heads, seq_len, head_dim)
+        # TODO: add support for attn.scale when we move to Torch 2.1
+        hidden_states = F.scaled_dot_product_attention(
+            query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
+        )
+
+        hidden_states = hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
+        hidden_states = hidden_states.to(query.dtype)
+        
+        # for ip
+        ip_key = self.to_k_ip(ip_hidden_states)
+        ip_value = self.to_v_ip(ip_hidden_states)
+        
+        ip_key = ip_key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+        ip_value = ip_value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
+
+        # the output of sdp = (batch, num_heads, seq_len, head_dim)
+        # TODO: add support for attn.scale when we move to Torch 2.1
+        ip_hidden_states = F.scaled_dot_product_attention(
+            query, ip_key, ip_value, attn_mask=None, dropout_p=0.0, is_causal=False
+        )
+        
+
+        ip_hidden_states = ip_hidden_states.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
+        ip_hidden_states = ip_hidden_states.to(query.dtype)
+        
+        hidden_states = hidden_states + self.scale * ip_hidden_states
+
+        # linear proj
+        hidden_states = attn.to_out[0](hidden_states) + self.lora_scale * self.to_out_lora(hidden_states)
+        # dropout
+        hidden_states = attn.to_out[1](hidden_states)
+
+        if input_ndim == 4:
+            hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)
+
+        if attn.residual_connection:
+            hidden_states = hidden_states + residual
+
+        hidden_states = hidden_states / attn.rescale_output_factor
+
+        return hidden_states

ip_adapter/custom_pipelines.py

@@ -0,0 +1,394 @@
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+import torch
+from diffusers import StableDiffusionXLPipeline
+from diffusers.pipelines.stable_diffusion_xl import StableDiffusionXLPipelineOutput
+from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl import rescale_noise_cfg
+
+from .utils import is_torch2_available
+
+if is_torch2_available():
+    from .attention_processor import IPAttnProcessor2_0 as IPAttnProcessor
+else:
+    from .attention_processor import IPAttnProcessor
+
+
+class StableDiffusionXLCustomPipeline(StableDiffusionXLPipeline):
+    def set_scale(self, scale):
+        for attn_processor in self.unet.attn_processors.values():
+            if isinstance(attn_processor, IPAttnProcessor):
+                attn_processor.scale = scale
+
+    @torch.no_grad()
+    def __call__(  # noqa: C901
+        self,
+        prompt: Optional[Union[str, List[str]]] = None,
+        prompt_2: Optional[Union[str, List[str]]] = None,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
+        num_inference_steps: int = 50,
+        denoising_end: Optional[float] = None,
+        guidance_scale: float = 5.0,
+        negative_prompt: Optional[Union[str, List[str]]] = None,
+        negative_prompt_2: Optional[Union[str, List[str]]] = None,
+        num_images_per_prompt: Optional[int] = 1,
+        eta: float = 0.0,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.FloatTensor] = None,
+        prompt_embeds: Optional[torch.FloatTensor] = None,
+        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
+        negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+        callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
+        callback_steps: int = 1,
+        cross_attention_kwargs: Optional[Dict[str, Any]] = None,
+        guidance_rescale: float = 0.0,
+        original_size: Optional[Tuple[int, int]] = None,
+        crops_coords_top_left: Tuple[int, int] = (0, 0),
+        target_size: Optional[Tuple[int, int]] = None,
+        negative_original_size: Optional[Tuple[int, int]] = None,
+        negative_crops_coords_top_left: Tuple[int, int] = (0, 0),
+        negative_target_size: Optional[Tuple[int, int]] = None,
+        control_guidance_start: float = 0.0,
+        control_guidance_end: float = 1.0,
+    ):
+        r"""
+        Function invoked when calling the pipeline for generation.
+
+        Args:
+            prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+                instead.
+            prompt_2 (`str` or `List[str]`, *optional*):
+                The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+                used in both text-encoders
+            height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
+                The height in pixels of the generated image. This is set to 1024 by default for the best results.
+                Anything below 512 pixels won't work well for
+                [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)
+                and checkpoints that are not specifically fine-tuned on low resolutions.
+            width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
+                The width in pixels of the generated image. This is set to 1024 by default for the best results.
+                Anything below 512 pixels won't work well for
+                [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)
+                and checkpoints that are not specifically fine-tuned on low resolutions.
+            num_inference_steps (`int`, *optional*, defaults to 50):
+                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+                expense of slower inference.
+            denoising_end (`float`, *optional*):
+                When specified, determines the fraction (between 0.0 and 1.0) of the total denoising process to be
+                completed before it is intentionally prematurely terminated. As a result, the returned sample will
+                still retain a substantial amount of noise as determined by the discrete timesteps selected by the
+                scheduler. The denoising_end parameter should ideally be utilized when this pipeline forms a part of a
+                "Mixture of Denoisers" multi-pipeline setup, as elaborated in [**Refining the Image
+                Output**](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl#refining-the-image-output)
+            guidance_scale (`float`, *optional*, defaults to 5.0):
+                Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
+                `guidance_scale` is defined as `w` of equation 2. of [Imagen
+                Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
+                1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
+                usually at the expense of lower image quality.
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation. If not defined, one has to pass
+                `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+                less than `1`).
+            negative_prompt_2 (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
+                `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
+            num_images_per_prompt (`int`, *optional*, defaults to 1):
+                The number of images to generate per prompt.
+            eta (`float`, *optional*, defaults to 0.0):
+                Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to
+                [`schedulers.DDIMScheduler`], will be ignored for others.
+            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+                One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
+                to make generation deterministic.
+            latents (`torch.FloatTensor`, *optional*):
+                Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor will ge generated by sampling using the supplied random `generator`.
+            prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+                provided, text embeddings will be generated from `prompt` input argument.
+            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+                argument.
+            pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+                If not provided, pooled text embeddings will be generated from `prompt` input argument.
+            negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
+                input argument.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generate image. Choose between
+                [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~pipelines.stable_diffusion_xl.StableDiffusionXLPipelineOutput`] instead
+                of a plain tuple.
+            callback (`Callable`, *optional*):
+                A function that will be called every `callback_steps` steps during inference. The function will be
+                called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
+            callback_steps (`int`, *optional*, defaults to 1):
+                The frequency at which the `callback` function will be called. If not specified, the callback will be
+                called at every step.
+            cross_attention_kwargs (`dict`, *optional*):
+                A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+                `self.processor` in
+                [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+            guidance_rescale (`float`, *optional*, defaults to 0.7):
+                Guidance rescale factor proposed by [Common Diffusion Noise Schedules and Sample Steps are
+                Flawed](https://arxiv.org/pdf/2305.08891.pdf) `guidance_scale` is defined as `φ` in equation 16. of
+                [Common Diffusion Noise Schedules and Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf).
+                Guidance rescale factor should fix overexposure when using zero terminal SNR.
+            original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+                If `original_size` is not the same as `target_size` the image will appear to be down- or upsampled.
+                `original_size` defaults to `(width, height)` if not specified. Part of SDXL's micro-conditioning as
+                explained in section 2.2 of
+                [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+            crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)):
+                `crops_coords_top_left` can be used to generate an image that appears to be "cropped" from the position
+                `crops_coords_top_left` downwards. Favorable, well-centered images are usually achieved by setting
+                `crops_coords_top_left` to (0, 0). Part of SDXL's micro-conditioning as explained in section 2.2 of
+                [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+            target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+                For most cases, `target_size` should be set to the desired height and width of the generated image. If
+                not specified it will default to `(width, height)`. Part of SDXL's micro-conditioning as explained in
+                section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+            negative_original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+                To negatively condition the generation process based on a specific image resolution. Part of SDXL's
+                micro-conditioning as explained in section 2.2 of
+                [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
+                information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
+            negative_crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)):
+                To negatively condition the generation process based on a specific crop coordinates. Part of SDXL's
+                micro-conditioning as explained in section 2.2 of
+                [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
+                information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
+            negative_target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+                To negatively condition the generation process based on a target image resolution. It should be as same
+                as the `target_size` for most cases. Part of SDXL's micro-conditioning as explained in section 2.2 of
+                [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
+                information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
+            control_guidance_start (`float`, *optional*, defaults to 0.0):
+                The percentage of total steps at which the ControlNet starts applying.
+            control_guidance_end (`float`, *optional*, defaults to 1.0):
+                The percentage of total steps at which the ControlNet stops applying.
+
+        Examples:
+
+        Returns:
+            [`~pipelines.stable_diffusion_xl.StableDiffusionXLPipelineOutput`] or `tuple`:
+            [`~pipelines.stable_diffusion_xl.StableDiffusionXLPipelineOutput`] if `return_dict` is True, otherwise a
+            `tuple`. When returning a tuple, the first element is a list with the generated images.
+        """
+        # 0. Default height and width to unet
+        height = height or self.default_sample_size * self.vae_scale_factor
+        width = width or self.default_sample_size * self.vae_scale_factor
+
+        original_size = original_size or (height, width)
+        target_size = target_size or (height, width)
+
+        # 1. Check inputs. Raise error if not correct
+        self.check_inputs(
+            prompt,
+            prompt_2,
+            height,
+            width,
+            callback_steps,
+            negative_prompt,
+            negative_prompt_2,
+            prompt_embeds,
+            negative_prompt_embeds,
+            pooled_prompt_embeds,
+            negative_pooled_prompt_embeds,
+        )
+
+        # 2. Define call parameters
+        if prompt is not None and isinstance(prompt, str):
+            batch_size = 1
+        elif prompt is not None and isinstance(prompt, list):
+            batch_size = len(prompt)
+        else:
+            batch_size = prompt_embeds.shape[0]
+
+        device = self._execution_device
+
+        # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+        # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
+        # corresponds to doing no classifier free guidance.
+        do_classifier_free_guidance = guidance_scale > 1.0
+
+        # 3. Encode input prompt
+        text_encoder_lora_scale = (
+            cross_attention_kwargs.get("scale", None) if cross_attention_kwargs is not None else None
+        )
+        (
+            prompt_embeds,
+            negative_prompt_embeds,
+            pooled_prompt_embeds,
+            negative_pooled_prompt_embeds,
+        ) = self.encode_prompt(
+            prompt=prompt,
+            prompt_2=prompt_2,
+            device=device,
+            num_images_per_prompt=num_images_per_prompt,
+            do_classifier_free_guidance=do_classifier_free_guidance,
+            negative_prompt=negative_prompt,
+            negative_prompt_2=negative_prompt_2,
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+            pooled_prompt_embeds=pooled_prompt_embeds,
+            negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
+            lora_scale=text_encoder_lora_scale,
+        )
+
+        # 4. Prepare timesteps
+        self.scheduler.set_timesteps(num_inference_steps, device=device)
+
+        timesteps = self.scheduler.timesteps
+
+        # 5. Prepare latent variables
+        num_channels_latents = self.unet.config.in_channels
+        latents = self.prepare_latents(
+            batch_size * num_images_per_prompt,
+            num_channels_latents,
+            height,
+            width,
+            prompt_embeds.dtype,
+            device,
+            generator,
+            latents,
+        )
+
+        # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+        extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+
+        # 7. Prepare added time ids & embeddings
+        add_text_embeds = pooled_prompt_embeds
+        if self.text_encoder_2 is None:
+            text_encoder_projection_dim = int(pooled_prompt_embeds.shape[-1])
+        else:
+            text_encoder_projection_dim = self.text_encoder_2.config.projection_dim
+
+        add_time_ids = self._get_add_time_ids(
+            original_size,
+            crops_coords_top_left,
+            target_size,
+            dtype=prompt_embeds.dtype,
+            text_encoder_projection_dim=text_encoder_projection_dim,
+        )
+        if negative_original_size is not None and negative_target_size is not None:
+            negative_add_time_ids = self._get_add_time_ids(
+                negative_original_size,
+                negative_crops_coords_top_left,
+                negative_target_size,
+                dtype=prompt_embeds.dtype,
+                text_encoder_projection_dim=text_encoder_projection_dim,
+            )
+        else:
+            negative_add_time_ids = add_time_ids
+
+        if do_classifier_free_guidance:
+            prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
+            add_text_embeds = torch.cat([negative_pooled_prompt_embeds, add_text_embeds], dim=0)
+            add_time_ids = torch.cat([negative_add_time_ids, add_time_ids], dim=0)
+
+        prompt_embeds = prompt_embeds.to(device)
+        add_text_embeds = add_text_embeds.to(device)
+        add_time_ids = add_time_ids.to(device).repeat(batch_size * num_images_per_prompt, 1)
+
+        # 8. Denoising loop
+        num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
+
+        # 7.1 Apply denoising_end
+        if denoising_end is not None and isinstance(denoising_end, float) and denoising_end > 0 and denoising_end < 1:
+            discrete_timestep_cutoff = int(
+                round(
+                    self.scheduler.config.num_train_timesteps
+                    - (denoising_end * self.scheduler.config.num_train_timesteps)
+                )
+            )
+            num_inference_steps = len(list(filter(lambda ts: ts >= discrete_timestep_cutoff, timesteps)))
+            timesteps = timesteps[:num_inference_steps]
+
+        # get init conditioning scale
+        for attn_processor in self.unet.attn_processors.values():
+            if isinstance(attn_processor, IPAttnProcessor):
+                conditioning_scale = attn_processor.scale
+                break
+
+        with self.progress_bar(total=num_inference_steps) as progress_bar:
+            for i, t in enumerate(timesteps):
+                if (i / len(timesteps) < control_guidance_start) or ((i + 1) / len(timesteps) > control_guidance_end):
+                    self.set_scale(0.0)
+                else:
+                    self.set_scale(conditioning_scale)
+
+                # expand the latents if we are doing classifier free guidance
+                latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
+
+                latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+
+                # predict the noise residual
+                added_cond_kwargs = {"text_embeds": add_text_embeds, "time_ids": add_time_ids}
+                noise_pred = self.unet(
+                    latent_model_input,
+                    t,
+                    encoder_hidden_states=prompt_embeds,
+                    cross_attention_kwargs=cross_attention_kwargs,
+                    added_cond_kwargs=added_cond_kwargs,
+                    return_dict=False,
+                )[0]
+
+                # perform guidance
+                if do_classifier_free_guidance:
+                    noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+                    noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+                if do_classifier_free_guidance and guidance_rescale > 0.0:
+                    # Based on 3.4. in https://arxiv.org/pdf/2305.08891.pdf
+                    noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=guidance_rescale)
+
+                # compute the previous noisy sample x_t -> x_t-1
+                latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+
+                # call the callback, if provided
+                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+                    progress_bar.update()
+                    if callback is not None and i % callback_steps == 0:
+                        callback(i, t, latents)
+
+        if not output_type == "latent":
+            # make sure the VAE is in float32 mode, as it overflows in float16
+            needs_upcasting = self.vae.dtype == torch.float16 and self.vae.config.force_upcast
+
+            if needs_upcasting:
+                self.upcast_vae()
+                latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)
+
+            image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]
+
+            # cast back to fp16 if needed
+            if needs_upcasting:
+                self.vae.to(dtype=torch.float16)
+        else:
+            image = latents
+
+        if output_type != "latent":
+            # apply watermark if available
+            if self.watermark is not None:
+                image = self.watermark.apply_watermark(image)
+
+            image = self.image_processor.postprocess(image, output_type=output_type)
+
+        # Offload all models
+        self.maybe_free_model_hooks()
+
+        if not return_dict:
+            return (image,)
+
+        return StableDiffusionXLPipelineOutput(images=image)

ip_adapter/ip_adapter.py

@@ -0,0 +1,417 @@
+import os
+from typing import List
+
+import torch
+from diffusers import StableDiffusionPipeline
+from diffusers.pipelines.controlnet import MultiControlNetModel
+from PIL import Image
+from safetensors import safe_open
+from transformers import CLIPImageProcessor, CLIPVisionModelWithProjection
+
+from .utils import is_torch2_available, get_generator
+
+if is_torch2_available():
+    from .attention_processor import (
+        AttnProcessor2_0 as AttnProcessor,
+    )
+    from .attention_processor import (
+        CNAttnProcessor2_0 as CNAttnProcessor,
+    )
+    from .attention_processor import (
+        IPAttnProcessor2_0 as IPAttnProcessor,
+    )
+else:
+    from .attention_processor import AttnProcessor, CNAttnProcessor, IPAttnProcessor
+from .resampler import Resampler
+
+
+class ImageProjModel(torch.nn.Module):
+    """Projection Model"""
+
+    def __init__(self, cross_attention_dim=1024, clip_embeddings_dim=1024, clip_extra_context_tokens=4):
+        super().__init__()
+
+        self.generator = None
+        self.cross_attention_dim = cross_attention_dim
+        self.clip_extra_context_tokens = clip_extra_context_tokens
+        self.proj = torch.nn.Linear(clip_embeddings_dim, self.clip_extra_context_tokens * cross_attention_dim)
+        self.norm = torch.nn.LayerNorm(cross_attention_dim)
+
+    def forward(self, image_embeds):
+        embeds = image_embeds
+        clip_extra_context_tokens = self.proj(embeds).reshape(
+            -1, self.clip_extra_context_tokens, self.cross_attention_dim
+        )
+        clip_extra_context_tokens = self.norm(clip_extra_context_tokens)
+        return clip_extra_context_tokens
+
+
+class MLPProjModel(torch.nn.Module):
+    """SD model with image prompt"""
+    def __init__(self, cross_attention_dim=1024, clip_embeddings_dim=1024):
+        super().__init__()
+        
+        self.proj = torch.nn.Sequential(
+            torch.nn.Linear(clip_embeddings_dim, clip_embeddings_dim),
+            torch.nn.GELU(),
+            torch.nn.Linear(clip_embeddings_dim, cross_attention_dim),
+            torch.nn.LayerNorm(cross_attention_dim)
+        )
+        
+    def forward(self, image_embeds):
+        clip_extra_context_tokens = self.proj(image_embeds)
+        return clip_extra_context_tokens
+
+
+class IPAdapter:
+    def __init__(self, sd_pipe, image_encoder_path, ip_ckpt, device, num_tokens=4):
+        self.device = device
+        self.image_encoder_path = image_encoder_path
+        self.ip_ckpt = ip_ckpt
+        self.num_tokens = num_tokens
+
+        self.pipe = sd_pipe.to(self.device)
+        self.set_ip_adapter()
+
+        # load image encoder
+        self.image_encoder = CLIPVisionModelWithProjection.from_pretrained(self.image_encoder_path).to(
+            self.device, dtype=torch.float16
+        )
+        self.clip_image_processor = CLIPImageProcessor()
+        # image proj model
+        self.image_proj_model = self.init_proj()
+
+        self.load_ip_adapter()
+
+    def init_proj(self):
+        image_proj_model = ImageProjModel(
+            cross_attention_dim=self.pipe.unet.config.cross_attention_dim,
+            clip_embeddings_dim=self.image_encoder.config.projection_dim,
+            clip_extra_context_tokens=self.num_tokens,
+        ).to(self.device, dtype=torch.float16)
+        return image_proj_model
+
+    def set_ip_adapter(self):
+        unet = self.pipe.unet
+        attn_procs = {}
+        for name in unet.attn_processors.keys():
+            cross_attention_dim = None if name.endswith("attn1.processor") else unet.config.cross_attention_dim
+            if name.startswith("mid_block"):
+                hidden_size = unet.config.block_out_channels[-1]
+            elif name.startswith("up_blocks"):
+                block_id = int(name[len("up_blocks.")])
+                hidden_size = list(reversed(unet.config.block_out_channels))[block_id]
+            elif name.startswith("down_blocks"):
+                block_id = int(name[len("down_blocks.")])
+                hidden_size = unet.config.block_out_channels[block_id]
+            if cross_attention_dim is None:
+                attn_procs[name] = AttnProcessor()
+            else:
+                attn_procs[name] = IPAttnProcessor(
+                    hidden_size=hidden_size,
+                    cross_attention_dim=cross_attention_dim,
+                    scale=1.0,
+                    num_tokens=self.num_tokens,
+                ).to(self.device, dtype=torch.float16)
+        unet.set_attn_processor(attn_procs)
+        if hasattr(self.pipe, "controlnet"):
+            if isinstance(self.pipe.controlnet, MultiControlNetModel):
+                for controlnet in self.pipe.controlnet.nets:
+                    controlnet.set_attn_processor(CNAttnProcessor(num_tokens=self.num_tokens))
+            else:
+                self.pipe.controlnet.set_attn_processor(CNAttnProcessor(num_tokens=self.num_tokens))
+
+    def load_ip_adapter(self):
+        if os.path.splitext(self.ip_ckpt)[-1] == ".safetensors":
+            state_dict = {"image_proj": {}, "ip_adapter": {}}
+            with safe_open(self.ip_ckpt, framework="pt", device="cpu") as f:
+                for key in f.keys():
+                    if key.startswith("image_proj."):
+                        state_dict["image_proj"][key.replace("image_proj.", "")] = f.get_tensor(key)
+                    elif key.startswith("ip_adapter."):
+                        state_dict["ip_adapter"][key.replace("ip_adapter.", "")] = f.get_tensor(key)
+        else:
+            state_dict = torch.load(self.ip_ckpt, map_location="cpu")
+        self.image_proj_model.load_state_dict(state_dict["image_proj"])
+        ip_layers = torch.nn.ModuleList(self.pipe.unet.attn_processors.values())
+        ip_layers.load_state_dict(state_dict["ip_adapter"])
+
+    @torch.inference_mode()
+    def get_image_embeds(self, pil_image=None, clip_image_embeds=None):
+        if pil_image is not None:
+            if isinstance(pil_image, Image.Image):
+                pil_image = [pil_image]
+            clip_image = self.clip_image_processor(images=pil_image, return_tensors="pt").pixel_values
+            clip_image_embeds = self.image_encoder(clip_image.to(self.device, dtype=torch.float16)).image_embeds
+        else:
+            clip_image_embeds = clip_image_embeds.to(self.device, dtype=torch.float16)
+        image_prompt_embeds = self.image_proj_model(clip_image_embeds)
+        uncond_image_prompt_embeds = self.image_proj_model(torch.zeros_like(clip_image_embeds))
+        return image_prompt_embeds, uncond_image_prompt_embeds
+
+    def set_scale(self, scale):
+        for attn_processor in self.pipe.unet.attn_processors.values():
+            if isinstance(attn_processor, IPAttnProcessor):
+                attn_processor.scale = scale
+
+    def generate(
+        self,
+        pil_image=None,
+        clip_image_embeds=None,
+        prompt=None,
+        negative_prompt=None,
+        scale=1.0,
+        num_samples=4,
+        seed=None,
+        guidance_scale=7.5,
+        num_inference_steps=30,
+        **kwargs,
+    ):
+        self.set_scale(scale)
+
+        if pil_image is not None:
+            num_prompts = 1 if isinstance(pil_image, Image.Image) else len(pil_image)
+        else:
+            num_prompts = clip_image_embeds.size(0)
+
+        if prompt is None:
+            prompt = "best quality, high quality"
+        if negative_prompt is None:
+            negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
+
+        if not isinstance(prompt, List):
+            prompt = [prompt] * num_prompts
+        if not isinstance(negative_prompt, List):
+            negative_prompt = [negative_prompt] * num_prompts
+
+        image_prompt_embeds, uncond_image_prompt_embeds = self.get_image_embeds(
+            pil_image=pil_image, clip_image_embeds=clip_image_embeds
+        )
+        bs_embed, seq_len, _ = image_prompt_embeds.shape
+        image_prompt_embeds = image_prompt_embeds.repeat(1, num_samples, 1)
+        image_prompt_embeds = image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
+        uncond_image_prompt_embeds = uncond_image_prompt_embeds.repeat(1, num_samples, 1)
+        uncond_image_prompt_embeds = uncond_image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
+
+        with torch.inference_mode():
+            prompt_embeds_, negative_prompt_embeds_ = self.pipe.encode_prompt(
+                prompt,
+                device=self.device,
+                num_images_per_prompt=num_samples,
+                do_classifier_free_guidance=True,
+                negative_prompt=negative_prompt,
+            )
+            prompt_embeds = torch.cat([prompt_embeds_, image_prompt_embeds], dim=1)
+            negative_prompt_embeds = torch.cat([negative_prompt_embeds_, uncond_image_prompt_embeds], dim=1)
+
+        generator = get_generator(seed, self.device)
+
+        images = self.pipe(
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+            guidance_scale=guidance_scale,
+            num_inference_steps=num_inference_steps,
+            generator=generator,
+            **kwargs,
+        ).images
+
+        return images
+
+
+class IPAdapterXL(IPAdapter):
+    """SDXL"""
+
+    def generate(
+        self,
+        pil_image,
+        prompt=None,
+        negative_prompt=None,
+        scale=1.0,
+        num_samples=4,
+        seed=None,
+        num_inference_steps=30,
+        **kwargs,
+    ):
+        self.set_scale(scale)
+
+        num_prompts = 1 if isinstance(pil_image, Image.Image) else len(pil_image)
+
+        if prompt is None:
+            prompt = "best quality, high quality"
+        if negative_prompt is None:
+            negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
+
+        if not isinstance(prompt, List):
+            prompt = [prompt] * num_prompts
+        if not isinstance(negative_prompt, List):
+            negative_prompt = [negative_prompt] * num_prompts
+
+        image_prompt_embeds, uncond_image_prompt_embeds = self.get_image_embeds(pil_image)
+        bs_embed, seq_len, _ = image_prompt_embeds.shape
+        image_prompt_embeds = image_prompt_embeds.repeat(1, num_samples, 1)
+        image_prompt_embeds = image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
+        uncond_image_prompt_embeds = uncond_image_prompt_embeds.repeat(1, num_samples, 1)
+        uncond_image_prompt_embeds = uncond_image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
+
+        with torch.inference_mode():
+            (
+                prompt_embeds,
+                negative_prompt_embeds,
+                pooled_prompt_embeds,
+                negative_pooled_prompt_embeds,
+            ) = self.pipe.encode_prompt(
+                prompt,
+                num_images_per_prompt=num_samples,
+                do_classifier_free_guidance=True,
+                negative_prompt=negative_prompt,
+            )
+            prompt_embeds = torch.cat([prompt_embeds, image_prompt_embeds], dim=1)
+            negative_prompt_embeds = torch.cat([negative_prompt_embeds, uncond_image_prompt_embeds], dim=1)
+
+        self.generator = get_generator(seed, self.device)
+        
+        images = self.pipe(
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+            pooled_prompt_embeds=pooled_prompt_embeds,
+            negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
+            num_inference_steps=num_inference_steps,
+            generator=self.generator,
+            **kwargs,
+        ).images
+
+        return images
+
+
+class IPAdapterPlus(IPAdapter):
+    """IP-Adapter with fine-grained features"""
+
+    def init_proj(self):
+        image_proj_model = Resampler(
+            dim=self.pipe.unet.config.cross_attention_dim,
+            depth=4,
+            dim_head=64,
+            heads=12,
+            num_queries=self.num_tokens,
+            embedding_dim=self.image_encoder.config.hidden_size,
+            output_dim=self.pipe.unet.config.cross_attention_dim,
+            ff_mult=4,
+        ).to(self.device, dtype=torch.float16)
+        return image_proj_model
+
+    @torch.inference_mode()
+    def get_image_embeds(self, pil_image=None, clip_image_embeds=None):
+        if isinstance(pil_image, Image.Image):
+            pil_image = [pil_image]
+        clip_image = self.clip_image_processor(images=pil_image, return_tensors="pt").pixel_values
+        clip_image = clip_image.to(self.device, dtype=torch.float16)
+        clip_image_embeds = self.image_encoder(clip_image, output_hidden_states=True).hidden_states[-2]
+        image_prompt_embeds = self.image_proj_model(clip_image_embeds)
+        uncond_clip_image_embeds = self.image_encoder(
+            torch.zeros_like(clip_image), output_hidden_states=True
+        ).hidden_states[-2]
+        uncond_image_prompt_embeds = self.image_proj_model(uncond_clip_image_embeds)
+        return image_prompt_embeds, uncond_image_prompt_embeds
+
+
+class IPAdapterFull(IPAdapterPlus):
+    """IP-Adapter with full features"""
+
+    def init_proj(self):
+        image_proj_model = MLPProjModel(
+            cross_attention_dim=self.pipe.unet.config.cross_attention_dim,
+            clip_embeddings_dim=self.image_encoder.config.hidden_size,
+        ).to(self.device, dtype=torch.float16)
+        return image_proj_model
+
+
+class IPAdapterPlusXL(IPAdapter):
+    """SDXL"""
+
+    def init_proj(self):
+        image_proj_model = Resampler(
+            dim=1280,
+            depth=4,
+            dim_head=64,
+            heads=20,
+            num_queries=self.num_tokens,
+            embedding_dim=self.image_encoder.config.hidden_size,
+            output_dim=self.pipe.unet.config.cross_attention_dim,
+            ff_mult=4,
+        ).to(self.device, dtype=torch.float16)
+        return image_proj_model
+
+    @torch.inference_mode()
+    def get_image_embeds(self, pil_image):
+        if isinstance(pil_image, Image.Image):
+            pil_image = [pil_image]
+        clip_image = self.clip_image_processor(images=pil_image, return_tensors="pt").pixel_values
+        clip_image = clip_image.to(self.device, dtype=torch.float16)
+        clip_image_embeds = self.image_encoder(clip_image, output_hidden_states=True).hidden_states[-2]
+        image_prompt_embeds = self.image_proj_model(clip_image_embeds)
+        uncond_clip_image_embeds = self.image_encoder(
+            torch.zeros_like(clip_image), output_hidden_states=True
+        ).hidden_states[-2]
+        uncond_image_prompt_embeds = self.image_proj_model(uncond_clip_image_embeds)
+        return image_prompt_embeds, uncond_image_prompt_embeds
+
+    def generate(
+        self,
+        pil_image,
+        prompt=None,
+        negative_prompt=None,
+        scale=1.0,
+        num_samples=4,
+        seed=None,
+        num_inference_steps=30,
+        **kwargs,
+    ):
+        self.set_scale(scale)
+
+        num_prompts = 1 if isinstance(pil_image, Image.Image) else len(pil_image)
+
+        if prompt is None:
+            prompt = "best quality, high quality"
+        if negative_prompt is None:
+            negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
+
+        if not isinstance(prompt, List):
+            prompt = [prompt] * num_prompts
+        if not isinstance(negative_prompt, List):
+            negative_prompt = [negative_prompt] * num_prompts
+
+        image_prompt_embeds, uncond_image_prompt_embeds = self.get_image_embeds(pil_image)
+        bs_embed, seq_len, _ = image_prompt_embeds.shape
+        image_prompt_embeds = image_prompt_embeds.repeat(1, num_samples, 1)
+        image_prompt_embeds = image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
+        uncond_image_prompt_embeds = uncond_image_prompt_embeds.repeat(1, num_samples, 1)
+        uncond_image_prompt_embeds = uncond_image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
+
+        with torch.inference_mode():
+            (
+                prompt_embeds,
+                negative_prompt_embeds,
+                pooled_prompt_embeds,
+                negative_pooled_prompt_embeds,
+            ) = self.pipe.encode_prompt(
+                prompt,
+                num_images_per_prompt=num_samples,
+                do_classifier_free_guidance=True,
+                negative_prompt=negative_prompt,
+            )
+            prompt_embeds = torch.cat([prompt_embeds, image_prompt_embeds], dim=1)
+            negative_prompt_embeds = torch.cat([negative_prompt_embeds, uncond_image_prompt_embeds], dim=1)
+
+        generator = get_generator(seed, self.device)
+
+        images = self.pipe(
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+            pooled_prompt_embeds=pooled_prompt_embeds,
+            negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
+            num_inference_steps=num_inference_steps,
+            generator=generator,
+            **kwargs,
+        ).images
+
+        return images

ip_adapter/ip_adapter_faceid.py

@@ -0,0 +1,542 @@
+import os
+from typing import List
+
+import torch
+from diffusers import StableDiffusionPipeline
+from diffusers.pipelines.controlnet import MultiControlNetModel
+from PIL import Image
+from safetensors import safe_open
+from transformers import CLIPImageProcessor, CLIPVisionModelWithProjection
+
+from .attention_processor_faceid import LoRAAttnProcessor, LoRAIPAttnProcessor
+from .utils import is_torch2_available, get_generator
+
+USE_DAFAULT_ATTN = False # should be True for visualization_attnmap
+if is_torch2_available() and (not USE_DAFAULT_ATTN):
+    from .attention_processor_faceid import (
+        LoRAAttnProcessor2_0 as LoRAAttnProcessor,
+    )
+    from .attention_processor_faceid import (
+        LoRAIPAttnProcessor2_0 as LoRAIPAttnProcessor,
+    )
+else:
+    from .attention_processor_faceid import LoRAAttnProcessor, LoRAIPAttnProcessor
+from .resampler import PerceiverAttention, FeedForward
+
+
+class FacePerceiverResampler(torch.nn.Module):
+    def __init__(
+        self,
+        *,
+        dim=768,
+        depth=4,
+        dim_head=64,
+        heads=16,
+        embedding_dim=1280,
+        output_dim=768,
+        ff_mult=4,
+    ):
+        super().__init__()
+        
+        self.proj_in = torch.nn.Linear(embedding_dim, dim)
+        self.proj_out = torch.nn.Linear(dim, output_dim)
+        self.norm_out = torch.nn.LayerNorm(output_dim)
+        self.layers = torch.nn.ModuleList([])
+        for _ in range(depth):
+            self.layers.append(
+                torch.nn.ModuleList(
+                    [
+                        PerceiverAttention(dim=dim, dim_head=dim_head, heads=heads),
+                        FeedForward(dim=dim, mult=ff_mult),
+                    ]
+                )
+            )
+
+    def forward(self, latents, x):
+        x = self.proj_in(x)
+        for attn, ff in self.layers:
+            latents = attn(x, latents) + latents
+            latents = ff(latents) + latents
+        latents = self.proj_out(latents)
+        return self.norm_out(latents)
+
+
+class MLPProjModel(torch.nn.Module):
+    def __init__(self, cross_attention_dim=768, id_embeddings_dim=512, num_tokens=4):
+        super().__init__()
+        
+        self.cross_attention_dim = cross_attention_dim
+        self.num_tokens = num_tokens
+        
+        self.proj = torch.nn.Sequential(
+            torch.nn.Linear(id_embeddings_dim, id_embeddings_dim*2),
+            torch.nn.GELU(),
+            torch.nn.Linear(id_embeddings_dim*2, cross_attention_dim*num_tokens),
+        )
+        self.norm = torch.nn.LayerNorm(cross_attention_dim)
+        
+    def forward(self, id_embeds):
+        x = self.proj(id_embeds)
+        x = x.reshape(-1, self.num_tokens, self.cross_attention_dim)
+        x = self.norm(x)
+        return x
+
+
+class ProjPlusModel(torch.nn.Module):
+    def __init__(self, cross_attention_dim=768, id_embeddings_dim=512, clip_embeddings_dim=1280, num_tokens=4):
+        super().__init__()
+        
+        self.cross_attention_dim = cross_attention_dim
+        self.num_tokens = num_tokens
+        
+        self.proj = torch.nn.Sequential(
+            torch.nn.Linear(id_embeddings_dim, id_embeddings_dim*2),
+            torch.nn.GELU(),
+            torch.nn.Linear(id_embeddings_dim*2, cross_attention_dim*num_tokens),
+        )
+        self.norm = torch.nn.LayerNorm(cross_attention_dim)
+        
+        self.perceiver_resampler = FacePerceiverResampler(
+            dim=cross_attention_dim,
+            depth=4,
+            dim_head=64,
+            heads=cross_attention_dim // 64,
+            embedding_dim=clip_embeddings_dim,
+            output_dim=cross_attention_dim,
+            ff_mult=4,
+        )
+        
+    def forward(self, id_embeds, clip_embeds, shortcut=False, scale=1.0):
+        
+        x = self.proj(id_embeds)
+        x = x.reshape(-1, self.num_tokens, self.cross_attention_dim)
+        x = self.norm(x)
+        out = self.perceiver_resampler(x, clip_embeds)
+        if shortcut:
+            out = x + scale * out
+        return out
+
+
+class IPAdapterFaceID:
+    def __init__(self, sd_pipe, ip_ckpt, device, lora_rank=128, num_tokens=4, torch_dtype=torch.float16):
+        self.device = device
+        self.ip_ckpt = ip_ckpt
+        self.lora_rank = lora_rank
+        self.num_tokens = num_tokens
+        self.torch_dtype = torch_dtype
+
+        self.pipe = sd_pipe.to(self.device)
+        self.set_ip_adapter()
+
+        # image proj model
+        self.image_proj_model = self.init_proj()
+
+        self.load_ip_adapter()
+
+    def init_proj(self):
+        image_proj_model = MLPProjModel(
+            cross_attention_dim=self.pipe.unet.config.cross_attention_dim,
+            id_embeddings_dim=512,
+            num_tokens=self.num_tokens,
+        ).to(self.device, dtype=self.torch_dtype)
+        return image_proj_model
+
+    def set_ip_adapter(self):
+        unet = self.pipe.unet
+        attn_procs = {}
+        for name in unet.attn_processors.keys():
+            cross_attention_dim = None if name.endswith("attn1.processor") else unet.config.cross_attention_dim
+            if name.startswith("mid_block"):
+                hidden_size = unet.config.block_out_channels[-1]
+            elif name.startswith("up_blocks"):
+                block_id = int(name[len("up_blocks.")])
+                hidden_size = list(reversed(unet.config.block_out_channels))[block_id]
+            elif name.startswith("down_blocks"):
+                block_id = int(name[len("down_blocks.")])
+                hidden_size = unet.config.block_out_channels[block_id]
+            if cross_attention_dim is None:
+                attn_procs[name] = LoRAAttnProcessor(
+                    hidden_size=hidden_size, cross_attention_dim=cross_attention_dim, rank=self.lora_rank,
+                ).to(self.device, dtype=self.torch_dtype)
+            else:
+                attn_procs[name] = LoRAIPAttnProcessor(
+                    hidden_size=hidden_size, cross_attention_dim=cross_attention_dim, scale=1.0, rank=self.lora_rank, num_tokens=self.num_tokens,
+                ).to(self.device, dtype=self.torch_dtype)
+        unet.set_attn_processor(attn_procs)
+
+    def load_ip_adapter(self):
+        if os.path.splitext(self.ip_ckpt)[-1] == ".safetensors":
+            state_dict = {"image_proj": {}, "ip_adapter": {}}
+            with safe_open(self.ip_ckpt, framework="pt", device="cpu") as f:
+                for key in f.keys():
+                    if key.startswith("image_proj."):
+                        state_dict["image_proj"][key.replace("image_proj.", "")] = f.get_tensor(key)
+                    elif key.startswith("ip_adapter."):
+                        state_dict["ip_adapter"][key.replace("ip_adapter.", "")] = f.get_tensor(key)
+        else:
+            state_dict = torch.load(self.ip_ckpt, map_location="cpu")
+        self.image_proj_model.load_state_dict(state_dict["image_proj"])
+        ip_layers = torch.nn.ModuleList(self.pipe.unet.attn_processors.values())
+        ip_layers.load_state_dict(state_dict["ip_adapter"])
+
+    @torch.inference_mode()
+    def get_image_embeds(self, faceid_embeds):
+        
+        faceid_embeds = faceid_embeds.to(self.device, dtype=self.torch_dtype)
+        image_prompt_embeds = self.image_proj_model(faceid_embeds)
+        uncond_image_prompt_embeds = self.image_proj_model(torch.zeros_like(faceid_embeds))
+        return image_prompt_embeds, uncond_image_prompt_embeds
+
+    def set_scale(self, scale):
+        for attn_processor in self.pipe.unet.attn_processors.values():
+            if isinstance(attn_processor, LoRAIPAttnProcessor):
+                attn_processor.scale = scale
+
+    def generate(
+        self,
+        faceid_embeds=None,
+        prompt=None,
+        negative_prompt=None,
+        scale=1.0,
+        num_samples=4,
+        seed=None,
+        guidance_scale=7.5,
+        num_inference_steps=30,
+        **kwargs,
+    ):
+        self.set_scale(scale)
+
+       
+        num_prompts = faceid_embeds.size(0)
+
+        if prompt is None:
+            prompt = "best quality, high quality"
+        if negative_prompt is None:
+            negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
+
+        if not isinstance(prompt, List):
+            prompt = [prompt] * num_prompts
+        if not isinstance(negative_prompt, List):
+            negative_prompt = [negative_prompt] * num_prompts
+
+        image_prompt_embeds, uncond_image_prompt_embeds = self.get_image_embeds(faceid_embeds)
+
+        bs_embed, seq_len, _ = image_prompt_embeds.shape
+        image_prompt_embeds = image_prompt_embeds.repeat(1, num_samples, 1)
+        image_prompt_embeds = image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
+        uncond_image_prompt_embeds = uncond_image_prompt_embeds.repeat(1, num_samples, 1)
+        uncond_image_prompt_embeds = uncond_image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
+
+        with torch.inference_mode():
+            prompt_embeds_, negative_prompt_embeds_ = self.pipe.encode_prompt(
+                prompt,
+                device=self.device,
+                num_images_per_prompt=num_samples,
+                do_classifier_free_guidance=True,
+                negative_prompt=negative_prompt,
+            )
+            prompt_embeds = torch.cat([prompt_embeds_, image_prompt_embeds], dim=1)
+            negative_prompt_embeds = torch.cat([negative_prompt_embeds_, uncond_image_prompt_embeds], dim=1)
+
+        generator = get_generator(seed, self.device)
+
+        images = self.pipe(
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+            guidance_scale=guidance_scale,
+            num_inference_steps=num_inference_steps,
+            generator=generator,
+            **kwargs,
+        ).images
+
+        return images
+
+
+class IPAdapterFaceIDPlus:
+    def __init__(self, sd_pipe, image_encoder_path, ip_ckpt, device, lora_rank=128, num_tokens=4, torch_dtype=torch.float16):
+        self.device = device
+        self.image_encoder_path = image_encoder_path
+        self.ip_ckpt = ip_ckpt
+        self.lora_rank = lora_rank
+        self.num_tokens = num_tokens
+        self.torch_dtype = torch_dtype
+
+        self.pipe = sd_pipe.to(self.device)
+        self.set_ip_adapter()
+
+        # load image encoder
+        self.image_encoder = CLIPVisionModelWithProjection.from_pretrained(self.image_encoder_path).to(
+            self.device, dtype=self.torch_dtype
+        )
+        self.clip_image_processor = CLIPImageProcessor()
+        # image proj model
+        self.image_proj_model = self.init_proj()
+
+        self.load_ip_adapter()
+
+    def init_proj(self):
+        image_proj_model = ProjPlusModel(
+            cross_attention_dim=self.pipe.unet.config.cross_attention_dim,
+            id_embeddings_dim=512,
+            clip_embeddings_dim=self.image_encoder.config.hidden_size,
+            num_tokens=self.num_tokens,
+        ).to(self.device, dtype=self.torch_dtype)
+        return image_proj_model
+
+    def set_ip_adapter(self):
+        unet = self.pipe.unet
+        attn_procs = {}
+        for name in unet.attn_processors.keys():
+            cross_attention_dim = None if name.endswith("attn1.processor") else unet.config.cross_attention_dim
+            if name.startswith("mid_block"):
+                hidden_size = unet.config.block_out_channels[-1]
+            elif name.startswith("up_blocks"):
+                block_id = int(name[len("up_blocks.")])
+                hidden_size = list(reversed(unet.config.block_out_channels))[block_id]
+            elif name.startswith("down_blocks"):
+                block_id = int(name[len("down_blocks.")])
+                hidden_size = unet.config.block_out_channels[block_id]
+            if cross_attention_dim is None:
+                attn_procs[name] = LoRAAttnProcessor(
+                    hidden_size=hidden_size, cross_attention_dim=cross_attention_dim, rank=self.lora_rank,
+                ).to(self.device, dtype=self.torch_dtype)
+            else:
+                attn_procs[name] = LoRAIPAttnProcessor(
+                    hidden_size=hidden_size, cross_attention_dim=cross_attention_dim, scale=1.0, rank=self.lora_rank, num_tokens=self.num_tokens,
+                ).to(self.device, dtype=self.torch_dtype)
+        unet.set_attn_processor(attn_procs)
+
+    def load_ip_adapter(self):
+        if os.path.splitext(self.ip_ckpt)[-1] == ".safetensors":
+            state_dict = {"image_proj": {}, "ip_adapter": {}}
+            with safe_open(self.ip_ckpt, framework="pt", device="cpu") as f:
+                for key in f.keys():
+                    if key.startswith("image_proj."):
+                        state_dict["image_proj"][key.replace("image_proj.", "")] = f.get_tensor(key)
+                    elif key.startswith("ip_adapter."):
+                        state_dict["ip_adapter"][key.replace("ip_adapter.", "")] = f.get_tensor(key)
+        else:
+            state_dict = torch.load(self.ip_ckpt, map_location="cpu")
+        self.image_proj_model.load_state_dict(state_dict["image_proj"])
+        ip_layers = torch.nn.ModuleList(self.pipe.unet.attn_processors.values())
+        ip_layers.load_state_dict(state_dict["ip_adapter"])
+
+    @torch.inference_mode()
+    def get_image_embeds(self, faceid_embeds, face_image, s_scale, shortcut):
+        if isinstance(face_image, Image.Image):
+            pil_image = [face_image]
+        clip_image = self.clip_image_processor(images=face_image, return_tensors="pt").pixel_values
+        clip_image = clip_image.to(self.device, dtype=self.torch_dtype)
+        clip_image_embeds = self.image_encoder(clip_image, output_hidden_states=True).hidden_states[-2]
+        uncond_clip_image_embeds = self.image_encoder(
+            torch.zeros_like(clip_image), output_hidden_states=True
+        ).hidden_states[-2]
+        
+        faceid_embeds = faceid_embeds.to(self.device, dtype=self.torch_dtype)
+        image_prompt_embeds = self.image_proj_model(faceid_embeds, clip_image_embeds, shortcut=shortcut, scale=s_scale)
+        uncond_image_prompt_embeds = self.image_proj_model(torch.zeros_like(faceid_embeds), uncond_clip_image_embeds, shortcut=shortcut, scale=s_scale)
+        return image_prompt_embeds, uncond_image_prompt_embeds
+
+    def set_scale(self, scale):
+        for attn_processor in self.pipe.unet.attn_processors.values():
+            if isinstance(attn_processor, LoRAIPAttnProcessor):
+                attn_processor.scale = scale
+
+    def generate(
+        self,
+        face_image=None,
+        faceid_embeds=None,
+        prompt=None,
+        negative_prompt=None,
+        scale=1.0,
+        num_samples=4,
+        seed=None,
+        guidance_scale=7.5,
+        num_inference_steps=30,
+        s_scale=1.0,
+        shortcut=False,
+        **kwargs,
+    ):
+        self.set_scale(scale)
+
+       
+        num_prompts = faceid_embeds.size(0)
+
+        if prompt is None:
+            prompt = "best quality, high quality"
+        if negative_prompt is None:
+            negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
+
+        if not isinstance(prompt, List):
+            prompt = [prompt] * num_prompts
+        if not isinstance(negative_prompt, List):
+            negative_prompt = [negative_prompt] * num_prompts
+
+        image_prompt_embeds, uncond_image_prompt_embeds = self.get_image_embeds(faceid_embeds, face_image, s_scale, shortcut)
+
+        bs_embed, seq_len, _ = image_prompt_embeds.shape
+        image_prompt_embeds = image_prompt_embeds.repeat(1, num_samples, 1)
+        image_prompt_embeds = image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
+        uncond_image_prompt_embeds = uncond_image_prompt_embeds.repeat(1, num_samples, 1)
+        uncond_image_prompt_embeds = uncond_image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
+
+        with torch.inference_mode():
+            prompt_embeds_, negative_prompt_embeds_ = self.pipe.encode_prompt(
+                prompt,
+                device=self.device,
+                num_images_per_prompt=num_samples,
+                do_classifier_free_guidance=True,
+                negative_prompt=negative_prompt,
+            )
+            prompt_embeds = torch.cat([prompt_embeds_, image_prompt_embeds], dim=1)
+            negative_prompt_embeds = torch.cat([negative_prompt_embeds_, uncond_image_prompt_embeds], dim=1)
+
+        generator = get_generator(seed, self.device)
+
+        images = self.pipe(
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+            guidance_scale=guidance_scale,
+            num_inference_steps=num_inference_steps,
+            generator=generator,
+            **kwargs,
+        ).images
+
+        return images
+
+
+class IPAdapterFaceIDXL(IPAdapterFaceID):
+    """SDXL"""
+
+    def generate(
+        self,
+        faceid_embeds=None,
+        prompt=None,
+        negative_prompt=None,
+        scale=1.0,
+        num_samples=4,
+        seed=None,
+        num_inference_steps=30,
+        **kwargs,
+    ):
+        self.set_scale(scale)
+
+        num_prompts = faceid_embeds.size(0)
+
+        if prompt is None:
+            prompt = "best quality, high quality"
+        if negative_prompt is None:
+            negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
+
+        if not isinstance(prompt, List):
+            prompt = [prompt] * num_prompts
+        if not isinstance(negative_prompt, List):
+            negative_prompt = [negative_prompt] * num_prompts
+
+        image_prompt_embeds, uncond_image_prompt_embeds = self.get_image_embeds(faceid_embeds)
+
+        bs_embed, seq_len, _ = image_prompt_embeds.shape
+        image_prompt_embeds = image_prompt_embeds.repeat(1, num_samples, 1)
+        image_prompt_embeds = image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
+        uncond_image_prompt_embeds = uncond_image_prompt_embeds.repeat(1, num_samples, 1)
+        uncond_image_prompt_embeds = uncond_image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
+
+        with torch.inference_mode():
+            (
+                prompt_embeds,
+                negative_prompt_embeds,
+                pooled_prompt_embeds,
+                negative_pooled_prompt_embeds,
+            ) = self.pipe.encode_prompt(
+                prompt,
+                num_images_per_prompt=num_samples,
+                do_classifier_free_guidance=True,
+                negative_prompt=negative_prompt,
+            )
+            prompt_embeds = torch.cat([prompt_embeds, image_prompt_embeds], dim=1)
+            negative_prompt_embeds = torch.cat([negative_prompt_embeds, uncond_image_prompt_embeds], dim=1)
+
+        generator = get_generator(seed, self.device)
+
+        images = self.pipe(
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+            pooled_prompt_embeds=pooled_prompt_embeds,
+            negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
+            num_inference_steps=num_inference_steps,
+            generator=generator,
+            **kwargs,
+        ).images
+
+        return images
+
+
+class IPAdapterFaceIDPlusXL(IPAdapterFaceIDPlus):
+    """SDXL"""
+
+    def generate(
+        self,
+        face_image=None,
+        faceid_embeds=None,
+        prompt=None,
+        negative_prompt=None,
+        scale=1.0,
+        num_samples=4,
+        seed=None,
+        guidance_scale=7.5,
+        num_inference_steps=30,
+        s_scale=1.0,
+        shortcut=True,
+        **kwargs,
+    ):
+        self.set_scale(scale)
+
+        num_prompts = faceid_embeds.size(0)
+
+        if prompt is None:
+            prompt = "best quality, high quality"
+        if negative_prompt is None:
+            negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
+
+        if not isinstance(prompt, List):
+            prompt = [prompt] * num_prompts
+        if not isinstance(negative_prompt, List):
+            negative_prompt = [negative_prompt] * num_prompts
+
+        image_prompt_embeds, uncond_image_prompt_embeds = self.get_image_embeds(faceid_embeds, face_image, s_scale, shortcut)
+
+        bs_embed, seq_len, _ = image_prompt_embeds.shape
+        image_prompt_embeds = image_prompt_embeds.repeat(1, num_samples, 1)
+        image_prompt_embeds = image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
+        uncond_image_prompt_embeds = uncond_image_prompt_embeds.repeat(1, num_samples, 1)
+        uncond_image_prompt_embeds = uncond_image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
+
+        with torch.inference_mode():
+            (
+                prompt_embeds,
+                negative_prompt_embeds,
+                pooled_prompt_embeds,
+                negative_pooled_prompt_embeds,
+            ) = self.pipe.encode_prompt(
+                prompt,
+                num_images_per_prompt=num_samples,
+                do_classifier_free_guidance=True,
+                negative_prompt=negative_prompt,
+            )
+            prompt_embeds = torch.cat([prompt_embeds, image_prompt_embeds], dim=1)
+            negative_prompt_embeds = torch.cat([negative_prompt_embeds, uncond_image_prompt_embeds], dim=1)
+
+        generator = get_generator(seed, self.device)
+
+        images = self.pipe(
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+            pooled_prompt_embeds=pooled_prompt_embeds,
+            negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
+            num_inference_steps=num_inference_steps,
+            generator=generator,
+            guidance_scale=guidance_scale,
+            **kwargs,
+        ).images
+
+        return images

ip_adapter/ip_adapter_faceid_separate.py

@@ -0,0 +1,547 @@
+import os
+from typing import List
+
+import torch
+from diffusers import StableDiffusionPipeline
+from diffusers.pipelines.controlnet import MultiControlNetModel
+from PIL import Image
+from safetensors import safe_open
+from transformers import CLIPImageProcessor, CLIPVisionModelWithProjection
+
+from .utils import is_torch2_available, get_generator
+
+USE_DAFAULT_ATTN = False # should be True for visualization_attnmap
+if is_torch2_available() and (not USE_DAFAULT_ATTN):
+    from .attention_processor import (
+        AttnProcessor2_0 as AttnProcessor,
+    )
+    from .attention_processor import (
+        IPAttnProcessor2_0 as IPAttnProcessor,
+    )
+else:
+    from .attention_processor import AttnProcessor, IPAttnProcessor
+from .resampler import PerceiverAttention, FeedForward
+
+
+class FacePerceiverResampler(torch.nn.Module):
+    def __init__(
+        self,
+        *,
+        dim=768,
+        depth=4,
+        dim_head=64,
+        heads=16,
+        embedding_dim=1280,
+        output_dim=768,
+        ff_mult=4,
+    ):
+        super().__init__()
+        
+        self.proj_in = torch.nn.Linear(embedding_dim, dim)
+        self.proj_out = torch.nn.Linear(dim, output_dim)
+        self.norm_out = torch.nn.LayerNorm(output_dim)
+        self.layers = torch.nn.ModuleList([])
+        for _ in range(depth):
+            self.layers.append(
+                torch.nn.ModuleList(
+                    [
+                        PerceiverAttention(dim=dim, dim_head=dim_head, heads=heads),
+                        FeedForward(dim=dim, mult=ff_mult),
+                    ]
+                )
+            )
+
+    def forward(self, latents, x):
+        x = self.proj_in(x)
+        for attn, ff in self.layers:
+            latents = attn(x, latents) + latents
+            latents = ff(latents) + latents
+        latents = self.proj_out(latents)
+        return self.norm_out(latents)
+
+
+class MLPProjModel(torch.nn.Module):
+    def __init__(self, cross_attention_dim=768, id_embeddings_dim=512, num_tokens=4):
+        super().__init__()
+        
+        self.cross_attention_dim = cross_attention_dim
+        self.num_tokens = num_tokens
+        
+        self.proj = torch.nn.Sequential(
+            torch.nn.Linear(id_embeddings_dim, id_embeddings_dim*2),
+            torch.nn.GELU(),
+            torch.nn.Linear(id_embeddings_dim*2, cross_attention_dim*num_tokens),
+        )
+        self.norm = torch.nn.LayerNorm(cross_attention_dim)
+        
+    def forward(self, id_embeds):
+        x = self.proj(id_embeds)
+        x = x.reshape(-1, self.num_tokens, self.cross_attention_dim)
+        x = self.norm(x)
+        return x
+
+
+class ProjPlusModel(torch.nn.Module):
+    def __init__(self, cross_attention_dim=768, id_embeddings_dim=512, clip_embeddings_dim=1280, num_tokens=4):
+        super().__init__()
+        
+        self.cross_attention_dim = cross_attention_dim
+        self.num_tokens = num_tokens
+        
+        self.proj = torch.nn.Sequential(
+            torch.nn.Linear(id_embeddings_dim, id_embeddings_dim*2),
+            torch.nn.GELU(),
+            torch.nn.Linear(id_embeddings_dim*2, cross_attention_dim*num_tokens),
+        )
+        self.norm = torch.nn.LayerNorm(cross_attention_dim)
+        
+        self.perceiver_resampler = FacePerceiverResampler(
+            dim=cross_attention_dim,
+            depth=4,
+            dim_head=64,
+            heads=cross_attention_dim // 64,
+            embedding_dim=clip_embeddings_dim,
+            output_dim=cross_attention_dim,
+            ff_mult=4,
+        )
+        
+    def forward(self, id_embeds, clip_embeds, shortcut=False, scale=1.0):
+        
+        x = self.proj(id_embeds)
+        x = x.reshape(-1, self.num_tokens, self.cross_attention_dim)
+        x = self.norm(x)
+        out = self.perceiver_resampler(x, clip_embeds)
+        if shortcut:
+            out = x + scale * out
+        return out
+
+
+class IPAdapterFaceID:
+    def __init__(self, sd_pipe, ip_ckpt, device, num_tokens=4, n_cond=1, torch_dtype=torch.float16):
+        self.device = device
+        self.ip_ckpt = ip_ckpt
+        self.num_tokens = num_tokens
+        self.n_cond = n_cond
+        self.torch_dtype = torch_dtype
+
+        self.pipe = sd_pipe.to(self.device)
+        self.set_ip_adapter()
+
+        # image proj model
+        self.image_proj_model = self.init_proj()
+
+        self.load_ip_adapter()
+
+    def init_proj(self):
+        image_proj_model = MLPProjModel(
+            cross_attention_dim=self.pipe.unet.config.cross_attention_dim,
+            id_embeddings_dim=512,
+            num_tokens=self.num_tokens,
+        ).to(self.device, dtype=self.torch_dtype)
+        return image_proj_model
+
+    def set_ip_adapter(self):
+        unet = self.pipe.unet
+        attn_procs = {}
+        for name in unet.attn_processors.keys():
+            cross_attention_dim = None if name.endswith("attn1.processor") else unet.config.cross_attention_dim
+            if name.startswith("mid_block"):
+                hidden_size = unet.config.block_out_channels[-1]
+            elif name.startswith("up_blocks"):
+                block_id = int(name[len("up_blocks.")])
+                hidden_size = list(reversed(unet.config.block_out_channels))[block_id]
+            elif name.startswith("down_blocks"):
+                block_id = int(name[len("down_blocks.")])
+                hidden_size = unet.config.block_out_channels[block_id]
+            if cross_attention_dim is None:
+                attn_procs[name] = AttnProcessor()
+            else:
+                attn_procs[name] = IPAttnProcessor(
+                    hidden_size=hidden_size, cross_attention_dim=cross_attention_dim, scale=1.0, num_tokens=self.num_tokens*self.n_cond,
+                ).to(self.device, dtype=self.torch_dtype)
+        unet.set_attn_processor(attn_procs)
+
+    def load_ip_adapter(self):
+        if os.path.splitext(self.ip_ckpt)[-1] == ".safetensors":
+            state_dict = {"image_proj": {}, "ip_adapter": {}}
+            with safe_open(self.ip_ckpt, framework="pt", device="cpu") as f:
+                for key in f.keys():
+                    if key.startswith("image_proj."):
+                        state_dict["image_proj"][key.replace("image_proj.", "")] = f.get_tensor(key)
+                    elif key.startswith("ip_adapter."):
+                        state_dict["ip_adapter"][key.replace("ip_adapter.", "")] = f.get_tensor(key)
+        else:
+            state_dict = torch.load(self.ip_ckpt, map_location="cpu")
+        self.image_proj_model.load_state_dict(state_dict["image_proj"])
+        ip_layers = torch.nn.ModuleList(self.pipe.unet.attn_processors.values())
+        ip_layers.load_state_dict(state_dict["ip_adapter"], strict=False)
+
+    @torch.inference_mode()
+    def get_image_embeds(self, faceid_embeds):
+
+        multi_face = False
+        if faceid_embeds.dim() == 3:
+            multi_face = True
+            b, n, c = faceid_embeds.shape
+            faceid_embeds = faceid_embeds.reshape(b*n, c)
+
+        faceid_embeds = faceid_embeds.to(self.device, dtype=self.torch_dtype)
+        image_prompt_embeds = self.image_proj_model(faceid_embeds)
+        uncond_image_prompt_embeds = self.image_proj_model(torch.zeros_like(faceid_embeds))
+        if multi_face:
+            c = image_prompt_embeds.size(-1)
+            image_prompt_embeds = image_prompt_embeds.reshape(b, -1, c)
+            uncond_image_prompt_embeds = uncond_image_prompt_embeds.reshape(b, -1, c)
+        
+        return image_prompt_embeds, uncond_image_prompt_embeds
+
+    def set_scale(self, scale):
+        for attn_processor in self.pipe.unet.attn_processors.values():
+            if isinstance(attn_processor, IPAttnProcessor):
+                attn_processor.scale = scale
+
+    def generate(
+        self,
+        faceid_embeds=None,
+        prompt=None,
+        negative_prompt=None,
+        scale=1.0,
+        num_samples=4,
+        seed=None,
+        guidance_scale=7.5,
+        num_inference_steps=30,
+        **kwargs,
+    ):
+        self.set_scale(scale)
+
+       
+        num_prompts = faceid_embeds.size(0)
+
+        if prompt is None:
+            prompt = "best quality, high quality"
+        if negative_prompt is None:
+            negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
+
+        if not isinstance(prompt, List):
+            prompt = [prompt] * num_prompts
+        if not isinstance(negative_prompt, List):
+            negative_prompt = [negative_prompt] * num_prompts
+
+        image_prompt_embeds, uncond_image_prompt_embeds = self.get_image_embeds(faceid_embeds)
+
+        bs_embed, seq_len, _ = image_prompt_embeds.shape
+        image_prompt_embeds = image_prompt_embeds.repeat(1, num_samples, 1)
+        image_prompt_embeds = image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
+        uncond_image_prompt_embeds = uncond_image_prompt_embeds.repeat(1, num_samples, 1)
+        uncond_image_prompt_embeds = uncond_image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
+
+        with torch.inference_mode():
+            prompt_embeds_, negative_prompt_embeds_ = self.pipe.encode_prompt(
+                prompt,
+                device=self.device,
+                num_images_per_prompt=num_samples,
+                do_classifier_free_guidance=True,
+                negative_prompt=negative_prompt,
+            )
+            prompt_embeds = torch.cat([prompt_embeds_, image_prompt_embeds], dim=1)
+            negative_prompt_embeds = torch.cat([negative_prompt_embeds_, uncond_image_prompt_embeds], dim=1)
+
+        generator = get_generator(seed, self.device)
+
+        images = self.pipe(
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+            guidance_scale=guidance_scale,
+            num_inference_steps=num_inference_steps,
+            generator=generator,
+            **kwargs,
+        ).images
+
+        return images
+
+
+class IPAdapterFaceIDPlus:
+    def __init__(self, sd_pipe, image_encoder_path, ip_ckpt, device, num_tokens=4, torch_dtype=torch.float16):
+        self.device = device
+        self.image_encoder_path = image_encoder_path
+        self.ip_ckpt = ip_ckpt
+        self.num_tokens = num_tokens
+        self.torch_dtype = torch_dtype
+
+        self.pipe = sd_pipe.to(self.device)
+        self.set_ip_adapter()
+
+        # load image encoder
+        self.image_encoder = CLIPVisionModelWithProjection.from_pretrained(self.image_encoder_path).to(
+            self.device, dtype=self.torch_dtype
+        )
+        self.clip_image_processor = CLIPImageProcessor()
+        # image proj model
+        self.image_proj_model = self.init_proj()
+
+        self.load_ip_adapter()
+
+    def init_proj(self):
+        image_proj_model = ProjPlusModel(
+            cross_attention_dim=self.pipe.unet.config.cross_attention_dim,
+            id_embeddings_dim=512,
+            clip_embeddings_dim=self.image_encoder.config.hidden_size,
+            num_tokens=self.num_tokens,
+        ).to(self.device, dtype=self.torch_dtype)
+        return image_proj_model
+
+    def set_ip_adapter(self):
+        unet = self.pipe.unet
+        attn_procs = {}
+        for name in unet.attn_processors.keys():
+            cross_attention_dim = None if name.endswith("attn1.processor") else unet.config.cross_attention_dim
+            if name.startswith("mid_block"):
+                hidden_size = unet.config.block_out_channels[-1]
+            elif name.startswith("up_blocks"):
+                block_id = int(name[len("up_blocks.")])
+                hidden_size = list(reversed(unet.config.block_out_channels))[block_id]
+            elif name.startswith("down_blocks"):
+                block_id = int(name[len("down_blocks.")])
+                hidden_size = unet.config.block_out_channels[block_id]
+            if cross_attention_dim is None:
+                attn_procs[name] = AttnProcessor()
+            else:
+                attn_procs[name] = IPAttnProcessor(
+                    hidden_size=hidden_size, cross_attention_dim=cross_attention_dim, scale=1.0, num_tokens=self.num_tokens,
+                ).to(self.device, dtype=self.torch_dtype)
+        unet.set_attn_processor(attn_procs)
+
+    def load_ip_adapter(self):
+        if os.path.splitext(self.ip_ckpt)[-1] == ".safetensors":
+            state_dict = {"image_proj": {}, "ip_adapter": {}}
+            with safe_open(self.ip_ckpt, framework="pt", device="cpu") as f:
+                for key in f.keys():
+                    if key.startswith("image_proj."):
+                        state_dict["image_proj"][key.replace("image_proj.", "")] = f.get_tensor(key)
+                    elif key.startswith("ip_adapter."):
+                        state_dict["ip_adapter"][key.replace("ip_adapter.", "")] = f.get_tensor(key)
+        else:
+            state_dict = torch.load(self.ip_ckpt, map_location="cpu")
+        self.image_proj_model.load_state_dict(state_dict["image_proj"])
+        ip_layers = torch.nn.ModuleList(self.pipe.unet.attn_processors.values())
+        ip_layers.load_state_dict(state_dict["ip_adapter"], strict=False)
+
+    @torch.inference_mode()
+    def get_image_embeds(self, faceid_embeds, face_image, s_scale, shortcut):
+        if isinstance(face_image, Image.Image):
+            pil_image = [face_image]
+        clip_image = self.clip_image_processor(images=face_image, return_tensors="pt").pixel_values
+        clip_image = clip_image.to(self.device, dtype=self.torch_dtype)
+        clip_image_embeds = self.image_encoder(clip_image, output_hidden_states=True).hidden_states[-2]
+        uncond_clip_image_embeds = self.image_encoder(
+            torch.zeros_like(clip_image), output_hidden_states=True
+        ).hidden_states[-2]
+        
+        faceid_embeds = faceid_embeds.to(self.device, dtype=self.torch_dtype)
+        image_prompt_embeds = self.image_proj_model(faceid_embeds, clip_image_embeds, shortcut=shortcut, scale=s_scale)
+        uncond_image_prompt_embeds = self.image_proj_model(torch.zeros_like(faceid_embeds), uncond_clip_image_embeds, shortcut=shortcut, scale=s_scale)
+        return image_prompt_embeds, uncond_image_prompt_embeds
+
+    def set_scale(self, scale):
+        for attn_processor in self.pipe.unet.attn_processors.values():
+            if isinstance(attn_processor, LoRAIPAttnProcessor):
+                attn_processor.scale = scale
+
+    def generate(
+        self,
+        face_image=None,
+        faceid_embeds=None,
+        prompt=None,
+        negative_prompt=None,
+        scale=1.0,
+        num_samples=4,
+        seed=None,
+        guidance_scale=7.5,
+        num_inference_steps=30,
+        s_scale=1.0,
+        shortcut=False,
+        **kwargs,
+    ):
+        self.set_scale(scale)
+
+       
+        num_prompts = faceid_embeds.size(0)
+
+        if prompt is None:
+            prompt = "best quality, high quality"
+        if negative_prompt is None:
+            negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
+
+        if not isinstance(prompt, List):
+            prompt = [prompt] * num_prompts
+        if not isinstance(negative_prompt, List):
+            negative_prompt = [negative_prompt] * num_prompts
+
+        image_prompt_embeds, uncond_image_prompt_embeds = self.get_image_embeds(faceid_embeds, face_image, s_scale, shortcut)
+
+        bs_embed, seq_len, _ = image_prompt_embeds.shape
+        image_prompt_embeds = image_prompt_embeds.repeat(1, num_samples, 1)
+        image_prompt_embeds = image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
+        uncond_image_prompt_embeds = uncond_image_prompt_embeds.repeat(1, num_samples, 1)
+        uncond_image_prompt_embeds = uncond_image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
+
+        with torch.inference_mode():
+            prompt_embeds_, negative_prompt_embeds_ = self.pipe.encode_prompt(
+                prompt,
+                device=self.device,
+                num_images_per_prompt=num_samples,
+                do_classifier_free_guidance=True,
+                negative_prompt=negative_prompt,
+            )
+            prompt_embeds = torch.cat([prompt_embeds_, image_prompt_embeds], dim=1)
+            negative_prompt_embeds = torch.cat([negative_prompt_embeds_, uncond_image_prompt_embeds], dim=1)
+
+        generator = get_generator(seed, self.device)
+
+        images = self.pipe(
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+            guidance_scale=guidance_scale,
+            num_inference_steps=num_inference_steps,
+            generator=generator,
+            **kwargs,
+        ).images
+
+        return images
+
+
+class IPAdapterFaceIDXL(IPAdapterFaceID):
+    """SDXL"""
+
+    def generate(
+        self,
+        faceid_embeds=None,
+        prompt=None,
+        negative_prompt=None,
+        scale=1.0,
+        num_samples=4,
+        seed=None,
+        num_inference_steps=30,
+        **kwargs,
+    ):
+        self.set_scale(scale)
+
+        num_prompts = faceid_embeds.size(0)
+
+        if prompt is None:
+            prompt = "best quality, high quality"
+        if negative_prompt is None:
+            negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
+
+        if not isinstance(prompt, List):
+            prompt = [prompt] * num_prompts
+        if not isinstance(negative_prompt, List):
+            negative_prompt = [negative_prompt] * num_prompts
+
+        image_prompt_embeds, uncond_image_prompt_embeds = self.get_image_embeds(faceid_embeds)
+
+        bs_embed, seq_len, _ = image_prompt_embeds.shape
+        image_prompt_embeds = image_prompt_embeds.repeat(1, num_samples, 1)
+        image_prompt_embeds = image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
+        uncond_image_prompt_embeds = uncond_image_prompt_embeds.repeat(1, num_samples, 1)
+        uncond_image_prompt_embeds = uncond_image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
+
+        with torch.inference_mode():
+            (
+                prompt_embeds,
+                negative_prompt_embeds,
+                pooled_prompt_embeds,
+                negative_pooled_prompt_embeds,
+            ) = self.pipe.encode_prompt(
+                prompt,
+                num_images_per_prompt=num_samples,
+                do_classifier_free_guidance=True,
+                negative_prompt=negative_prompt,
+            )
+            prompt_embeds = torch.cat([prompt_embeds, image_prompt_embeds], dim=1)
+            negative_prompt_embeds = torch.cat([negative_prompt_embeds, uncond_image_prompt_embeds], dim=1)
+
+        generator = get_generator(seed, self.device)
+
+        images = self.pipe(
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+            pooled_prompt_embeds=pooled_prompt_embeds,
+            negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
+            num_inference_steps=num_inference_steps,
+            generator=generator,
+            **kwargs,
+        ).images
+
+        return images
+
+
+class IPAdapterFaceIDPlusXL(IPAdapterFaceIDPlus):
+    """SDXL"""
+
+    def generate(
+        self,
+        face_image=None,
+        faceid_embeds=None,
+        prompt=None,
+        negative_prompt=None,
+        scale=1.0,
+        num_samples=4,
+        seed=None,
+        guidance_scale=7.5,
+        num_inference_steps=30,
+        s_scale=1.0,
+        shortcut=True,
+        **kwargs,
+    ):
+        self.set_scale(scale)
+
+        num_prompts = faceid_embeds.size(0)
+
+        if prompt is None:
+            prompt = "best quality, high quality"
+        if negative_prompt is None:
+            negative_prompt = "monochrome, lowres, bad anatomy, worst quality, low quality"
+
+        if not isinstance(prompt, List):
+            prompt = [prompt] * num_prompts
+        if not isinstance(negative_prompt, List):
+            negative_prompt = [negative_prompt] * num_prompts
+
+        image_prompt_embeds, uncond_image_prompt_embeds = self.get_image_embeds(faceid_embeds, face_image, s_scale, shortcut)
+
+        bs_embed, seq_len, _ = image_prompt_embeds.shape
+        image_prompt_embeds = image_prompt_embeds.repeat(1, num_samples, 1)
+        image_prompt_embeds = image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
+        uncond_image_prompt_embeds = uncond_image_prompt_embeds.repeat(1, num_samples, 1)
+        uncond_image_prompt_embeds = uncond_image_prompt_embeds.view(bs_embed * num_samples, seq_len, -1)
+
+        with torch.inference_mode():
+            (
+                prompt_embeds,
+                negative_prompt_embeds,
+                pooled_prompt_embeds,
+                negative_pooled_prompt_embeds,
+            ) = self.pipe.encode_prompt(
+                prompt,
+                num_images_per_prompt=num_samples,
+                do_classifier_free_guidance=True,
+                negative_prompt=negative_prompt,
+            )
+            prompt_embeds = torch.cat([prompt_embeds, image_prompt_embeds], dim=1)
+            negative_prompt_embeds = torch.cat([negative_prompt_embeds, uncond_image_prompt_embeds], dim=1)
+
+        generator = get_generator(seed, self.device)
+
+        images = self.pipe(
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+            pooled_prompt_embeds=pooled_prompt_embeds,
+            negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
+            num_inference_steps=num_inference_steps,
+            generator=generator,
+            guidance_scale=guidance_scale,
+            **kwargs,
+        ).images
+
+        return images

ip_adapter/resampler.py

@@ -0,0 +1,158 @@
+# modified from https://github.com/mlfoundations/open_flamingo/blob/main/open_flamingo/src/helpers.py
+# and https://github.com/lucidrains/imagen-pytorch/blob/main/imagen_pytorch/imagen_pytorch.py
+
+import math
+
+import torch
+import torch.nn as nn
+from einops import rearrange
+from einops.layers.torch import Rearrange
+
+
+# FFN
+def FeedForward(dim, mult=4):
+    inner_dim = int(dim * mult)
+    return nn.Sequential(
+        nn.LayerNorm(dim),
+        nn.Linear(dim, inner_dim, bias=False),
+        nn.GELU(),
+        nn.Linear(inner_dim, dim, bias=False),
+    )
+
+
+def reshape_tensor(x, heads):
+    bs, length, width = x.shape
+    # (bs, length, width) --> (bs, length, n_heads, dim_per_head)
+    x = x.view(bs, length, heads, -1)
+    # (bs, length, n_heads, dim_per_head) --> (bs, n_heads, length, dim_per_head)
+    x = x.transpose(1, 2)
+    # (bs, n_heads, length, dim_per_head) --> (bs*n_heads, length, dim_per_head)
+    x = x.reshape(bs, heads, length, -1)
+    return x
+
+
+class PerceiverAttention(nn.Module):
+    def __init__(self, *, dim, dim_head=64, heads=8):
+        super().__init__()
+        self.scale = dim_head**-0.5
+        self.dim_head = dim_head
+        self.heads = heads
+        inner_dim = dim_head * heads
+
+        self.norm1 = nn.LayerNorm(dim)
+        self.norm2 = nn.LayerNorm(dim)
+
+        self.to_q = nn.Linear(dim, inner_dim, bias=False)
+        self.to_kv = nn.Linear(dim, inner_dim * 2, bias=False)
+        self.to_out = nn.Linear(inner_dim, dim, bias=False)
+
+    def forward(self, x, latents):
+        """
+        Args:
+            x (torch.Tensor): image features
+                shape (b, n1, D)
+            latent (torch.Tensor): latent features
+                shape (b, n2, D)
+        """
+        x = self.norm1(x)
+        latents = self.norm2(latents)
+
+        b, l, _ = latents.shape
+
+        q = self.to_q(latents)
+        kv_input = torch.cat((x, latents), dim=-2)
+        k, v = self.to_kv(kv_input).chunk(2, dim=-1)
+
+        q = reshape_tensor(q, self.heads)
+        k = reshape_tensor(k, self.heads)
+        v = reshape_tensor(v, self.heads)
+
+        # attention
+        scale = 1 / math.sqrt(math.sqrt(self.dim_head))
+        weight = (q * scale) @ (k * scale).transpose(-2, -1)  # More stable with f16 than dividing afterwards
+        weight = torch.softmax(weight.float(), dim=-1).type(weight.dtype)
+        out = weight @ v
+
+        out = out.permute(0, 2, 1, 3).reshape(b, l, -1)
+
+        return self.to_out(out)
+
+
+class Resampler(nn.Module):
+    def __init__(
+        self,
+        dim=1024,
+        depth=8,
+        dim_head=64,
+        heads=16,
+        num_queries=8,
+        embedding_dim=768,
+        output_dim=1024,
+        ff_mult=4,
+        max_seq_len: int = 257,  # CLIP tokens + CLS token
+        apply_pos_emb: bool = False,
+        num_latents_mean_pooled: int = 0,  # number of latents derived from mean pooled representation of the sequence
+    ):
+        super().__init__()
+        self.pos_emb = nn.Embedding(max_seq_len, embedding_dim) if apply_pos_emb else None
+
+        self.latents = nn.Parameter(torch.randn(1, num_queries, dim) / dim**0.5)
+
+        self.proj_in = nn.Linear(embedding_dim, dim)
+
+        self.proj_out = nn.Linear(dim, output_dim)
+        self.norm_out = nn.LayerNorm(output_dim)
+
+        self.to_latents_from_mean_pooled_seq = (
+            nn.Sequential(
+                nn.LayerNorm(dim),
+                nn.Linear(dim, dim * num_latents_mean_pooled),
+                Rearrange("b (n d) -> b n d", n=num_latents_mean_pooled),
+            )
+            if num_latents_mean_pooled > 0
+            else None
+        )
+
+        self.layers = nn.ModuleList([])
+        for _ in range(depth):
+            self.layers.append(
+                nn.ModuleList(
+                    [
+                        PerceiverAttention(dim=dim, dim_head=dim_head, heads=heads),
+                        FeedForward(dim=dim, mult=ff_mult),
+                    ]
+                )
+            )
+
+    def forward(self, x):
+        if self.pos_emb is not None:
+            n, device = x.shape[1], x.device
+            pos_emb = self.pos_emb(torch.arange(n, device=device))
+            x = x + pos_emb
+
+        latents = self.latents.repeat(x.size(0), 1, 1)
+
+        x = self.proj_in(x)
+
+        if self.to_latents_from_mean_pooled_seq:
+            meanpooled_seq = masked_mean(x, dim=1, mask=torch.ones(x.shape[:2], device=x.device, dtype=torch.bool))
+            meanpooled_latents = self.to_latents_from_mean_pooled_seq(meanpooled_seq)
+            latents = torch.cat((meanpooled_latents, latents), dim=-2)
+
+        for attn, ff in self.layers:
+            latents = attn(x, latents) + latents
+            latents = ff(latents) + latents
+
+        latents = self.proj_out(latents)
+        return self.norm_out(latents)
+
+
+def masked_mean(t, *, dim, mask=None):
+    if mask is None:
+        return t.mean(dim=dim)
+
+    denom = mask.sum(dim=dim, keepdim=True)
+    mask = rearrange(mask, "b n -> b n 1")
+    masked_t = t.masked_fill(~mask, 0.0)
+
+    return masked_t.sum(dim=dim) / denom.clamp(min=1e-5)

ip_adapter/test_resampler.py

@@ -0,0 +1,44 @@
+import torch
+from resampler import Resampler
+from transformers import CLIPVisionModel
+
+BATCH_SIZE = 2
+OUTPUT_DIM = 1280
+NUM_QUERIES = 8
+NUM_LATENTS_MEAN_POOLED = 4  # 0 for no mean pooling (previous behavior)
+APPLY_POS_EMB = True  # False for no positional embeddings (previous behavior)
+IMAGE_ENCODER_NAME_OR_PATH = "laion/CLIP-ViT-H-14-laion2B-s32B-b79K"
+
+
+def main():
+    image_encoder = CLIPVisionModel.from_pretrained(IMAGE_ENCODER_NAME_OR_PATH)
+    embedding_dim = image_encoder.config.hidden_size
+    print(f"image_encoder hidden size: ", embedding_dim)
+
+    image_proj_model = Resampler(
+        dim=1024,
+        depth=2,
+        dim_head=64,
+        heads=16,
+        num_queries=NUM_QUERIES,
+        embedding_dim=embedding_dim,
+        output_dim=OUTPUT_DIM,
+        ff_mult=2,
+        max_seq_len=257,
+        apply_pos_emb=APPLY_POS_EMB,
+        num_latents_mean_pooled=NUM_LATENTS_MEAN_POOLED,
+    )
+
+    dummy_images = torch.randn(BATCH_SIZE, 3, 224, 224)
+    with torch.no_grad():
+        image_embeds = image_encoder(dummy_images, output_hidden_states=True).hidden_states[-2]
+    print("image_embds shape: ", image_embeds.shape)
+
+    with torch.no_grad():
+        ip_tokens = image_proj_model(image_embeds)
+    print("ip_tokens shape:", ip_tokens.shape)
+    assert ip_tokens.shape == (BATCH_SIZE, NUM_QUERIES + NUM_LATENTS_MEAN_POOLED, OUTPUT_DIM)
+
+
+if __name__ == "__main__":
+    main()

ip_adapter/utils.py

@@ -0,0 +1,93 @@
+import torch
+import torch.nn.functional as F
+import numpy as np
+from PIL import Image
+
+attn_maps = {}
+def hook_fn(name):
+    def forward_hook(module, input, output):
+        if hasattr(module.processor, "attn_map"):
+            attn_maps[name] = module.processor.attn_map
+            del module.processor.attn_map
+
+    return forward_hook
+
+def register_cross_attention_hook(unet):
+    for name, module in unet.named_modules():
+        if name.split('.')[-1].startswith('attn2'):
+            module.register_forward_hook(hook_fn(name))
+
+    return unet
+
+def upscale(attn_map, target_size):
+    attn_map = torch.mean(attn_map, dim=0)
+    attn_map = attn_map.permute(1,0)
+    temp_size = None
+
+    for i in range(0,5):
+        scale = 2 ** i
+        if ( target_size[0] // scale ) * ( target_size[1] // scale) == attn_map.shape[1]*64:
+            temp_size = (target_size[0]//(scale*8), target_size[1]//(scale*8))
+            break
+
+    assert temp_size is not None, "temp_size cannot is None"
+
+    attn_map = attn_map.view(attn_map.shape[0], *temp_size)
+
+    attn_map = F.interpolate(
+        attn_map.unsqueeze(0).to(dtype=torch.float32),
+        size=target_size,
+        mode='bilinear',
+        align_corners=False
+    )[0]
+
+    attn_map = torch.softmax(attn_map, dim=0)
+    return attn_map
+def get_net_attn_map(image_size, batch_size=2, instance_or_negative=False, detach=True):
+
+    idx = 0 if instance_or_negative else 1
+    net_attn_maps = []
+
+    for name, attn_map in attn_maps.items():
+        attn_map = attn_map.cpu() if detach else attn_map
+        attn_map = torch.chunk(attn_map, batch_size)[idx].squeeze()
+        attn_map = upscale(attn_map, image_size) 
+        net_attn_maps.append(attn_map) 
+
+    net_attn_maps = torch.mean(torch.stack(net_attn_maps,dim=0),dim=0)
+
+    return net_attn_maps
+
+def attnmaps2images(net_attn_maps):
+
+    #total_attn_scores = 0
+    images = []
+
+    for attn_map in net_attn_maps:
+        attn_map = attn_map.cpu().numpy()
+        #total_attn_scores += attn_map.mean().item()
+
+        normalized_attn_map = (attn_map - np.min(attn_map)) / (np.max(attn_map) - np.min(attn_map)) * 255
+        normalized_attn_map = normalized_attn_map.astype(np.uint8)
+        #print("norm: ", normalized_attn_map.shape)
+        image = Image.fromarray(normalized_attn_map)
+
+        #image = fix_save_attn_map(attn_map)
+        images.append(image)
+
+    #print(total_attn_scores)
+    return images
+def is_torch2_available():
+    return hasattr(F, "scaled_dot_product_attention")
+
+def get_generator(seed, device):
+
+    if seed is not None:
+        if isinstance(seed, list):
+            generator = [torch.Generator(device).manual_seed(seed_item) for seed_item in seed]
+        else:
+            generator = torch.Generator(device).manual_seed(seed)
+    else:
+        generator = None
+
+    return generator

models/image_encoder/config.json

@@ -0,0 +1,23 @@
+{
+  "_name_or_path": "./image_encoder",
+  "architectures": [
+    "CLIPVisionModelWithProjection"
+  ],
+  "attention_dropout": 0.0,
+  "dropout": 0.0,
+  "hidden_act": "gelu",
+  "hidden_size": 1280,
+  "image_size": 224,
+  "initializer_factor": 1.0,
+  "initializer_range": 0.02,
+  "intermediate_size": 5120,
+  "layer_norm_eps": 1e-05,
+  "model_type": "clip_vision_model",
+  "num_attention_heads": 16,
+  "num_channels": 3,
+  "num_hidden_layers": 32,
+  "patch_size": 14,
+  "projection_dim": 1024,
+  "torch_dtype": "float16",
+  "transformers_version": "4.28.0.dev0"
+}

requirements.txt

@@ -0,0 +1,10 @@
+diffusers
+rembg
+einops==0.7.0
+transformers==4.27.4
+opencv-python==4.7.0.68
+gradio==4.15.0
+accelerate==0.26.1
+timm==0.6.12
+torch==2.0.1
+torchvision==0.15.2

run_batch.py

@@ -0,0 +1,93 @@
+from diffusers import StableDiffusionXLControlNetInpaintPipeline, ControlNetModel
+from rembg import remove
+from PIL import Image
+import torch
+from ip_adapter import IPAdapterXL
+from ip_adapter.utils import register_cross_attention_hook, get_net_attn_map, attnmaps2images
+from PIL import Image, ImageChops
+from PIL import ImageEnhance
+import numpy as np
+import glob
+
+def image_grid(imgs, rows, cols):
+    assert len(imgs) == rows*cols
+
+    w, h = imgs[0].size
+    grid = Image.new('RGB', size=(cols*w, rows*h))
+    grid_w, grid_h = grid.size
+    
+    for i, img in enumerate(imgs):
+        grid.paste(img, box=(i%cols*w, i//cols*h))
+    return grid
+
+base_model_path = "stabilityai/stable-diffusion-xl-base-1.0"
+image_encoder_path = "models/image_encoder"
+ip_ckpt = "sdxl_models/ip-adapter_sdxl_vit-h.bin"
+controlnet_path = "diffusers/controlnet-depth-sdxl-1.0"
+device = "cuda"
+
+torch.cuda.empty_cache()
+
+# load SDXL pipeline
+controlnet = ControlNetModel.from_pretrained(controlnet_path, variant="fp16", use_safetensors=True, torch_dtype=torch.float16).to(device)
+pipe = StableDiffusionXLControlNetInpaintPipeline.from_pretrained(
+    base_model_path,
+    controlnet=controlnet,
+    use_safetensors=True,
+    torch_dtype=torch.float16,
+    add_watermarker=False,
+).to(device)
+pipe.unet = register_cross_attention_hook(pipe.unet)
+
+ip_model = IPAdapterXL(pipe, image_encoder_path, ip_ckpt, device)
+
+
+
+textures = [tex.split('/')[-1].replace('.png', '') for tex in glob.glob('demo_assets/material_exemplars/*.png')]
+objs = [obj.split('/')[-1].replace('.png', '') for obj in glob.glob('demo_assets/input_imgs/*.png')]
+
+for texture in textures:
+    for obj in objs:
+        target_image_path = 'demo_assets/input_imgs/' + obj + '.png'  # Replace with your image path
+        target_image = Image.open(target_image_path).convert('RGB')
+        rm_bg = remove(target_image)
+        # output.save(output_path)
+        target_mask = rm_bg.convert("RGB").point(lambda x: 0 if x < 1 else 255).convert('L').convert('RGB')# Convert mask to grayscale
+
+        # Ensure mask is the same size as image
+
+        # mask = ImageChops.invert(mask)
+        # Generate random noise for the size of the image
+        noise = np.random.randint(0, 256, target_image.size + (3,), dtype=np.uint8)
+        noise_image = Image.fromarray(noise)
+        mask_target_img = ImageChops.lighter(target_image, target_mask)
+        invert_target_mask = ImageChops.invert(target_mask)
+
+
+        gray_target_image = target_image.convert('L').convert('RGB')
+        gray_target_image = ImageEnhance.Brightness(gray_target_image)
+
+        # Adjust brightness
+        # The factor 1.0 means original brightness, greater than 1.0 makes the image brighter. Adjust this if the image is too dim
+        factor = 1.0  # Try adjusting this to get the desired brightness
+
+        gray_target_image = gray_target_image.enhance(factor)
+        grayscale_img = ImageChops.darker(gray_target_image, target_mask)
+        img_black_mask = ImageChops.darker(target_image, invert_target_mask)
+        grayscale_init_img = ImageChops.lighter(img_black_mask, grayscale_img)
+        init_img = grayscale_init_img
+
+        ip_image = Image.open("demo_assets/material_exemplars/" + texture + ".png")
+        np_image = np.array(Image.open('demo_assets/depths/' + obj + '.png'))
+        
+        np_image = (np_image / 256).astype('uint8')
+
+        depth_map = Image.fromarray(np_image).resize((1024,1024))
+
+        init_img = init_img.resize((1024,1024))
+        mask = target_mask.resize((1024, 1024))
+        
+        num_samples = 1
+        images = ip_model.generate(pil_image=ip_image, image=init_img, control_image=depth_map, mask_image=mask, controlnet_conditioning_scale=0.9, num_samples=num_samples, num_inference_steps=30, seed=42)
+        images[0].save('demo_assets/output_images/' + obj + '_' + texture + '.png' )
+

sdxl_models/image_encoder/config.json

@@ -0,0 +1,81 @@
+{
+  "architectures": [
+    "CLIPVisionModelWithProjection"
+  ],
+  "_name_or_path": "",
+  "add_cross_attention": false,
+  "architectures": null,
+  "attention_dropout": 0.0,
+  "bad_words_ids": null,
+  "begin_suppress_tokens": null,
+  "bos_token_id": null,
+  "chunk_size_feed_forward": 0,
+  "cross_attention_hidden_size": null,
+  "decoder_start_token_id": null,
+  "diversity_penalty": 0.0,
+  "do_sample": false,
+  "dropout": 0.0,
+  "early_stopping": false,
+  "encoder_no_repeat_ngram_size": 0,
+  "eos_token_id": null,
+  "exponential_decay_length_penalty": null,
+  "finetuning_task": null,
+  "forced_bos_token_id": null,
+  "forced_eos_token_id": null,
+  "hidden_act": "gelu",
+  "hidden_size": 1664,
+  "id2label": {
+    "0": "LABEL_0",
+    "1": "LABEL_1"
+      },
+  "image_size": 224,
+  "initializer_factor": 1.0,
+  "initializer_range": 0.02,
+  "intermediate_size": 8192,
+  "is_decoder": false,
+  "is_encoder_decoder": false,
+  "label2id": {
+    "LABEL_0": 0,
+    "LABEL_1": 1
+      },
+  "layer_norm_eps": 1e-05,
+  "length_penalty": 1.0,
+  "max_length": 20,
+  "min_length": 0,
+  "model_type": "clip_vision_model",
+  "no_repeat_ngram_size": 0,
+  "num_attention_heads": 16,
+  "num_beam_groups": 1,
+  "num_beams": 1,
+  "num_channels": 3,
+  "num_hidden_layers": 48,
+  "num_return_sequences": 1,
+  "output_attentions": false,
+  "output_hidden_states": false,
+  "output_scores": false,
+  "pad_token_id": null,
+  "patch_size": 14,
+  "prefix": null,
+  "problem_type": null,
+  "pruned_heads": {},
+  "remove_invalid_values": false,
+  "repetition_penalty": 1.0,
+  "return_dict": true,
+  "return_dict_in_generate": false,
+  "sep_token_id": null,
+  "suppress_tokens": null,
+  "task_specific_params": null,
+  "temperature": 1.0,
+  "tf_legacy_loss": false,
+  "tie_encoder_decoder": false,
+  "tie_word_embeddings": true,
+  "tokenizer_class": null,
+  "top_k": 50,
+  "top_p": 1.0,
+  "torch_dtype": null,
+  "torchscript": false,
+  "transformers_version": "4.24.0",
+  "typical_p": 1.0,
+  "use_bfloat16": false,
+  "projection_dim": 1280
+}

visualization.py

@@ -0,0 +1,18 @@
+import glob
+from html4vision import Col, imagetable
+
+textures = [tex.split('/')[-1].replace('.png', '') for tex in glob.glob('demo_assets/material_exemplars/*.png')]
+objs = [obj.split('/')[-1].replace('.png', '') for obj in glob.glob('demo_assets/input_imgs/*.png')]
+
+# Generate first column as input images for reference
+cols = []
+cols.append(Col('img', '',[''] + ['demo_assets/input_imgs/' + obj + '.png' for obj in objs ] ))
+
+# Generate each column of results
+for texture in textures:
+    cur_col =['demo_assets/material_exemplars/' + texture + '.png']
+    for obj in objs:
+        cur_col.append('demo_assets/output_images/' + texture + '_' + obj + '.png')
+    cols.append(Col('img', texture, cur_col))
+
+imagetable(cols)