app.py

import gradio as gr
import timm
import torch
from PIL import Image
import requests
from io import BytesIO
import numpy as np
from pytorch_grad_cam import GradCAM, HiResCAM, ScoreCAM, GradCAMPlusPlus, AblationCAM, XGradCAM, EigenCAM, FullGrad
from pytorch_grad_cam.utils.model_targets import ClassifierOutputTarget, get_target_layer
from pytorch_grad_cam.utils.image import show_cam_on_image
from timm.data import create_transform

# List of available timm models
MODELS = timm.list_models()

# List of available GradCAM methods
CAM_METHODS = {
    "GradCAM": GradCAM,
    "HiResCAM": HiResCAM,
    "ScoreCAM": ScoreCAM,
    "GradCAM++": GradCAMPlusPlus,
    "AblationCAM": AblationCAM,
    "XGradCAM": XGradCAM,
    "EigenCAM": EigenCAM,
    "FullGrad": FullGrad
}

def load_model(model_name):
    model = timm.create_model(model_name, pretrained=True)
    model.eval()
    return model

def process_image(image_path, model):
    if image_path.startswith('http'):
        response = requests.get(image_path)
        image = Image.open(BytesIO(response.content))
    else:
        image = Image.open(image_path)
    
    config = model.pretrained_cfg
    transform = create_transform(
        input_size=config['input_size'],
        crop_pct=config['crop_pct'],
        mean=config['mean'],
        std=config['std'],
        interpolation=config['interpolation'],
        is_training=False
    )
    
    tensor = transform(image).unsqueeze(0)
    return tensor

def get_cam_image(model, image, target_layer, cam_method):
    cam = CAM_METHODS[cam_method](model=model, target_layers=[target_layer], use_cuda=torch.cuda.is_available())
    grayscale_cam = cam(input_tensor=image)
    
    config = model.pretrained_cfg
    mean = torch.tensor(config['mean']).view(3, 1, 1)
    std = torch.tensor(config['std']).view(3, 1, 1)
    rgb_img = (image.squeeze(0) * std + mean).permute(1, 2, 0).cpu().numpy()
    rgb_img = np.clip(rgb_img, 0, 1)
    
    cam_image = show_cam_on_image(rgb_img, grayscale_cam[0, :], use_rgb=True)
    return Image.fromarray(cam_image)

def get_feature_info(model):
    if hasattr(model, 'feature_info'):
        return [f['module'] for f in model.feature_info]
    else:
        return []

def explain_image(model_name, image_path, cam_method, feature_module):
    model = load_model(model_name)
    image = process_image(image_path, model)
    
    if feature_module:
        target_layer = get_target_layer(model, feature_module)
        print(f"Using feature module: {feature_module}")
    else:
        # Fallback to the last feature module or last convolutional layer
        feature_info = get_feature_info(model)
        if feature_info:
            target_layer = get_target_layer(model, feature_info[-1])
            print(f"Using last feature module: {feature_info[-1]}")
        else:
            # Fallback to finding last convolutional layer
            for name, module in reversed(list(model.named_modules())):
                if isinstance(module, torch.nn.Conv2d):
                    target_layer = module
                    print(f"Fallback: Using last convolutional layer: {name}")
                    break
    
    if target_layer is None:
        raise ValueError("Could not find a suitable target layer.")
    
    cam_image = get_cam_image(model, image, target_layer, cam_method)
    return cam_image

def update_feature_modules(model_name):
    model = load_model(model_name)
    feature_modules = get_feature_info(model)
    return gr.Dropdown.update(choices=feature_modules, value=feature_modules[-1] if feature_modules else None)

iface = gr.Interface(
    fn=explain_image,
    inputs=[
        gr.Dropdown(choices=MODELS, label="Select Model"),
        gr.Image(type="filepath", label="Upload Image"),
        gr.Dropdown(choices=list(CAM_METHODS.keys()), label="Select CAM Method"),
        gr.Dropdown(label="Select Feature Module (optional)")
    ],
    outputs=gr.Image(type="pil", label="Explained Image"),
    title="Explainable AI with timm models",
    description="Upload an image, select a model, CAM method, and optionally a specific feature module to visualize the explanation.",
    allow_flagging="never"
)

iface.load(update_feature_modules, inputs=[iface.inputs[0]], outputs=[iface.inputs[3]])

iface.launch()