Upload folder using huggingface_hub

README.md

@@ -5,10 +5,10 @@ colorFrom: green
 colorTo: red
 sdk: gradio
 python_version: 3.10.13
-sdk_version: 3.48.0
+sdk_version: 4.36.1
 app_file: app.py
 pinned: false
 license: mit
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -1,5 +1,5 @@
-import spaces
 import gradio as gr
+import sys
 import torch
 from omegaconf import OmegaConf
 from PIL import Image
@@ -9,10 +9,18 @@ import cv2
 import numpy as np
 import argparse
 
+# Parse command line arguments
+parser = argparse.ArgumentParser()
+parser.add_argument("--config", type=str, default="config/inference_config.yaml", help="Path to the config file")
+parser.add_argument("--share", action="store_true", help="Share the interface if provided")
+args = parser.parse_args()
+
+device = 'cuda' if torch.cuda.is_available() else 'cpu'
+
 # Load configuration and models
-config = OmegaConf.load("config/inference_config.yaml")
+config = OmegaConf.load(args.config)
 sd_pipeline = StableDiffusionInpaintPipeline.from_pretrained(
-    "runwayml/stable-diffusion-inpainting", torch_dtype=torch.float32
+    "runwayml/stable-diffusion-inpainting", safety_checker=None, torch_dtype=torch.float32
 )
 clipaway = CLIPAway(
     sd_pipe=sd_pipeline, 
@@ -21,7 +29,7 @@ clipaway = CLIPAway(
     alpha_clip_path=config.alpha_clip_ckpt_pth, 
     config=config, 
     alpha_clip_id=config.alpha_clip_id, 
-    device="cpu", 
+    device=device, 
     num_tokens=4
 )
 
@@ -43,7 +51,7 @@ def remove_obj(image, uploaded_mask, seed):
     image_pil, sketched_mask = image["image"], image["mask"]
     mask = dilate_mask(combine_masks(uploaded_mask, sketched_mask))
     seed = int(seed)
-    latents = torch.randn((1, 4, 64, 64), generator=torch.Generator().manual_seed(seed)).to("cpu")
+    latents = torch.randn((1, 4, 64, 64), generator=torch.Generator().manual_seed(seed)).to(device)
     final_image = clipaway.generate(
         prompt=[""], scale=1, seed=seed,
         pil_image=[image_pil], alpha=[mask], strength=1, latents=latents
@@ -52,29 +60,29 @@ def remove_obj(image, uploaded_mask, seed):
 
 # Define example data
 examples = [
-    ["gradio_examples/images/1.jpg", "gradio_examples/masks/1.png", 42],
-    ["gradio_examples/images/2.jpg", "gradio_examples/masks/2.png", 42],
-    ["gradio_examples/images/3.jpg", "gradio_examples/masks/3.png", 464],
+    ["assets/gradio_examples/images/1.jpg", "assets/gradio_examples/masks/1.png", 42],
+    ["assets/gradio_examples/images/2.jpg", "assets/gradio_examples/masks/2.png", 42],
+    ["assets/gradio_examples/images/3.jpg", "assets/gradio_examples/masks/3.png", 2024],
 ]
 
+# Define the Gradio interface
 with gr.Blocks() as demo:
     gr.Markdown("<h1 style='text-align:center'>CLIPAway: Harmonizing Focused Embeddings for Removing Objects via Diffusion Models</h1>")
     gr.Markdown("""
         <div style='display:flex; justify-content:center; align-items:center;'>
-            <a href='https://arxiv.org/abs/2406.09368' style="margin-right:10px;">Paper</a> |
+            <a href='https://arxiv.org/abs/2406.09368' style="margin:10px;">Paper</a> |
             <a href='https://yigitekin.github.io/CLIPAway/' style="margin:10px;">Project Website</a> |
-            <a href='https://github.com/YigitEkin/CLIPAway' style="margin-left:10px;">GitHub</a>
+            <a href='https://github.com/YigitEkin/CLIPAway' style="margin:10px;">GitHub</a>
         </div>
     """)
     gr.Markdown("""
             This application allows you to remove objects from images using the CLIPAway method with diffusion models.
             To use this tool:
-            1. Upload an image. (NOTE: We expect a 512x512 image, if you upload a different size, it will be resized to 512x512 which can affect the results.)
-            2. Upload a pre-defined mask if you have one. (If you don't have a mask, and want to sketch one, 
-            we have provided a gradio demo in our github repository. <br/> Unfortunately, we cannot provide it here due to the compatibility issues with zerogpu.)
-            3. Set the seed for reproducibility (default is 42).
-            4. Click 'Remove Object' to process the image.
-            5. The result will be displayed on the right side.
+            1. Upload an image.
+            2. Either Sketch a mask over the object you want to remove or upload a pre-defined mask if you have one.
+            4. Set the seed for reproducibility (default is 42).
+            5. Click 'Remove Object' to process the image.
+            6. The result will be displayed on the right side.
             Note: The mask should be a binary image where the object to be removed is white and the background is black.
     """)
     
@@ -89,10 +97,10 @@ with gr.Blocks() as demo:
     
     process_button.click(
         fn=remove_obj,
-        inputs=[image_input, seed_input],
+        inputs=[image_input, uploaded_mask, seed_input],
         outputs=result_image
     )
 
+# Launch the interface with caching
 
-
-demo.launch(share=True)
+demo.launch(share=True)

model/clip_away.py

@@ -38,8 +38,8 @@ class ImageProjModel(torch.nn.Module):
         return clip_extra_context_tokens
 
 class CLIPAway:
-    def __init__(self, sd_pipe, image_encoder_path, ip_ckpt, alpha_clip_path, config, alpha_clip_id="ViT-L/14", device="cuda", num_tokens=4):
-        super().__init__()
+    def __init__(self, sd_pipe, image_encoder_path, ip_ckpt, alpha_clip_path, config, device, alpha_clip_id="ViT-L/14", num_tokens=4):
+        super().__init__()        
         self.device = device
         self.ipadapter_image_encoder_path = image_encoder_path
         self.ipadapter_ckpt = ip_ckpt

requirements.txt

@@ -15,5 +15,4 @@ transformers==4.39.3
 git+https://github.com/openai/CLIP.git
 git+https://github.com/tencent-ailab/IP-Adapter.git
 git+https://github.com/SunzeY/AlphaCLIP.git
-loralib
-gradio==4.44.1
+loralib