app.py

import gradio as gr
import numpy as np
import torch
from PIL import Image
from skimage.feature import graycomatrix, graycoprops
from torchvision import transforms

# Load the model
model = torch.jit.load("SuSy.pt")

def process_image(image):
    # Set Parameters
    top_k_patches = 5
    patch_size = 224

    # Get the image dimensions
    width, height = image.size

    # Calculate the number of patches
    num_patches_x = width // patch_size
    num_patches_y = height // patch_size

    # Divide the image in patches
    patches = np.zeros((num_patches_x * num_patches_y, patch_size, patch_size, 3), dtype=np.uint8)
    for i in range(num_patches_x):
        for j in range(num_patches_y):
            x = i * patch_size
            y = j * patch_size
            patch = image.crop((x, y, x + patch_size, y + patch_size))
            patches[i * num_patches_y + j] = np.array(patch)

    # Compute the most relevant patches (optional)
    dissimilarity_scores = []
    for patch in patches:
        transform_patch = transforms.Compose([transforms.PILToTensor(), transforms.Grayscale()])
        grayscale_patch = transform_patch(Image.fromarray(patch)).squeeze(0)
        glcm = graycomatrix(grayscale_patch, [5], [0], 256, symmetric=True, normed=True)
        dissimilarity_scores.append(graycoprops(glcm, "contrast")[0, 0])

    # Sort patch indices by their dissimilarity score
    sorted_indices = np.argsort(dissimilarity_scores)[::-1]

    # Extract top k patches and convert them to tensor
    top_patches = patches[sorted_indices[:top_k_patches]]
    top_patches = torch.from_numpy(np.transpose(top_patches, (0, 3, 1, 2))) / 255.0

    # Predict patches
    model.eval()
    with torch.no_grad():
        preds = model(top_patches)

    # Process results
    classes = ['Authentic', 'DALL·E 3', 'Stable Diffusion 1.x', 'MJ V5/V6', 'MJ V1/V2', 'Stable Diffusion XL']
    mean_probs = preds.mean(dim=0).numpy()
    
    # Create a dictionary of class probabilities
    class_probs = {cls: prob for cls, prob in zip(classes, mean_probs)}
    
    # Sort probabilities in descending order
    sorted_probs = dict(sorted(class_probs.items(), key=lambda item: item[1], reverse=True))

    return sorted_probs

# Define Gradio interface
iface = gr.Interface(
    fn=process_image,
    inputs=gr.Image(type="pil"),
    outputs=gr.Label(num_top_classes=6),
    title="SuSy: Synthetic Image Detector",
    description="""
    <table style="border-collapse: collapse; border: none; padding: 20px;">
    <tr style="border: none;">
        <td style="border: none; vertical-align: top; padding-right: 30px; padding-left: 30px;">
        <img src="https://cdn-uploads.huggingface.co/production/uploads/62f7a16192950415b637e201/NobqlpFbFkTyBi1LsT9JE.png" alt="SuSy Logo" width="120" style="margin-bottom: 10px;">
        </td>
        <td style="border: none; vertical-align: top; padding: 10px;">
        <p style="margin-bottom: 15px;">Detect synthetic images with SuSy! SuSy can distinguish between authentic images and those generated by DALL·E, Midjourney and Stable Diffusion.</p>
        <p style="margin-top: 15px;">Learn more about SuSy: <a href="https://arxiv.org/abs/2409.14128">Present and Future Generalization of Synthetic Image Detectors</a></p>
        <p style="margin-top: 15px;">
            Enter the SuSy-verse!
            <a href="https://huggingface.co/HPAI-BSC/SuSy">Model</a> |
            <a href="https://github.com/HPAI-BSC/SuSy">Code</a> |
            <a href="https://huggingface.co/datasets/HPAI-BSC/SuSy-Dataset">Dataset</a>
        </p>
        </td>
    </tr>
    </table>
    """,
    examples=[
        ["example_authentic.jpg"],
        ["example_dalle3.jpg"],
        ["example_mjv5.jpg"],
        ["example_sdxl.jpg"]
    ]
)

# Launch the interface
iface.launch()