vision-diffmask / app.py
Orpheous1
probs
5dc90b6
raw
history blame
3.63 kB
import sys
sys.path.insert(0, './code')
from datamodules.transformations import UnNest
from models.interpretation import ImageInterpretationNet
from transformers import ViTFeatureExtractor, ViTForImageClassification
from utils.plot import smoothen, draw_mask_on_image, draw_heatmap_on_image
import gradio as gr
import numpy as np
import torch
import seaborn as sns
import matplotlib.pyplot as plt
# Load Vision Transformer
hf_model = "tanlq/vit-base-patch16-224-in21k-finetuned-cifar10"
hf_model_imagenet = "google/vit-base-patch16-224"
vit = ViTForImageClassification.from_pretrained(hf_model)
vit_imagenet = ViTForImageClassification.from_pretrained(hf_model_imagenet)
vit.eval()
vit_imagenet.eval()
# Load Feature Extractor
feature_extractor = ViTFeatureExtractor.from_pretrained(hf_model, return_tensors="pt")
feature_extractor_imagenet = ViTFeatureExtractor.from_pretrained(hf_model_imagenet, return_tensors="pt")
feature_extractor = UnNest(feature_extractor)
feature_extractor_imagenet = UnNest(feature_extractor_imagenet)
# Load Vision DiffMask
diffmask = ImageInterpretationNet.load_from_checkpoint('checkpoints/diffmask.ckpt')
diffmask.set_vision_transformer(vit)
diffmask_imagenet = ImageInterpretationNet.load_from_checkpoint('checkpoints/diffmask_imagenet.ckpt')
diffmask_imagenet.set_vision_transformer(vit_imagenet)
diffmask.eval()
diffmask_imagenet.eval()
# Define mask plotting functions
def draw_mask(image, mask):
return draw_mask_on_image(image, smoothen(mask))\
.permute(1, 2, 0)\
.clip(0, 1)\
.numpy()
def draw_heatmap(image, mask):
return draw_heatmap_on_image(image, smoothen(mask))\
.permute(1, 2, 0)\
.clip(0, 1)\
.numpy()
# Define callable method for the demo
def get_mask(image, model_name: str):
if image is None:
return None, None
if model_name == 'DiffMask-CiFAR-10':
diffmask_model = diffmask
elif model_name == 'DiffMask-ImageNet':
diffmask_model = diffmask_imagenet
image = torch.from_numpy(image).permute(2, 0, 1).float() / 255
dm_image = feature_extractor(image).unsqueeze(0)
dm_out = diffmask_model.get_mask(dm_image)
mask = dm_out["mask"][0].detach()
logits = dm_out["logits"][0].detach().softmax(dim=-1)
logits_orig = dm_out["logits_orig"][0].detach().softmax(dim=-1)
# fig, ax = plt.subplots(nrows=1, ncols=1, figsize=(10, 10))
# sns.displot(logits_orig.cpu().numpy().flatten(), kind="kde", label="Original", ax=ax)
top5logits_orig = logits_orig.topk(5, dim=-1)
idx = top5logits_orig.indices
# keep the top 5 classes from the indices of the top 5 logits
top5logits_orig = top5logits_orig.values
top5logits = logits[idx]
pred = dm_out["pred_class"][0].detach()
pred = diffmask_model.model.config.id2label[pred.item()]
masked_img = draw_mask(image, mask)
heatmap = draw_heatmap(image, mask)
orig_probs = {diffmask_model.model.config.id2label[i]: top5logits_orig[i].item() for i in range(5)}
pred_probs = {diffmask_model.model.config.id2label[i]: top5logits[i].item() for i in range(5)}
return np.hstack((masked_img, heatmap)), pred, orig_probs, pred_probs
# Launch demo interface
gr.Interface(
get_mask,
inputs=[gr.inputs.Image(label="Input", shape=(224, 224), source="upload", type="numpy"),
gr.inputs.Dropdown(["DiffMask-CiFAR-10", "DiffMask-ImageNet"])],
outputs=[gr.outputs.Image(label="Output"), gr.outputs.Label(label="Prediction"),
gr.Label(label="Original Probabilities"), gr.Label(label="Predicted Probabilities")],
title="Vision DiffMask Demo",
live=True,
).launch()