import gradio as gr import torch import torch.nn as nn from torchvision.models import resnet18, ResNet18_Weights from torchvision import transforms as T from PIL import Image class SiameseNetwork(nn.Module): def __init__(self): super().__init__() self.resnet = resnet18(weights=ResNet18_Weights.DEFAULT) self.resnet.conv1 = nn.Conv2d(1, 64, kernel_size=9, stride=2, padding=4, bias=False) self.resnet = nn.Sequential(*list(self.resnet.children())[:-1]) self.fc = nn.Sequential( nn.Linear(512 * 2, 256), nn.ReLU(), nn.Dropout(0.3), nn.Linear(256, 1), nn.Sigmoid() ) def forward_once(self, x): x = self.resnet(x) return x.view(x.size(0), -1) def forward(self, x1, x2): f1 = self.forward_once(x1) f2 = self.forward_once(x2) return self.fc(torch.cat([f1, f2], dim=1)) transform = T.Compose([ T.Resize((64, 64)), T.Grayscale(num_output_channels=1), T.ToTensor(), T.Normalize((0.5,), (0.5,)) ]) def preprocess_image(image): image = image.convert("L") tensor = transform(image).unsqueeze(0) return tensor.to(device) device = torch.device("cuda" if torch.cuda.is_available() else "cpu") model = SiameseNetwork().to(device) model.load_state_dict(torch.load("model_weights.pt", map_location=device)) model.eval() def predict_similarity(image1, image2): try: img1 = preprocess_image(image1) img2 = preprocess_image(image2) with torch.no_grad(): similarity = model(img1, img2).item() match_str = "Yes" if similarity > 0.5 else "No" return f"Similarity Score (0-1): {similarity:.4f}\nMatch: {match_str}" except Exception as e: return f"Error: {str(e)}" demo = gr.Interface( fn=predict_similarity, inputs=[gr.Image(type="pil"), gr.Image(type="pil")], outputs="text", title="MNIST Siamese Similarity", description="Upload two digit images to compare similarity." ) if __name__ == "__main__": demo.launch()