import torch import torch.nn as nn import torch.nn.functional as F from torchvision import transforms from PIL import Image import numpy as np import matplotlib.pyplot as plt import gradio as gr import os # ========================= # 1. DEVICE CONFIG # ========================= DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu") MODEL_PATH = "sharp_edge_unet.pth" # Path to trained model # ========================= # 2. DEFINE SHARP UNET # ========================= class SharpUNet(nn.Module): def __init__(self): super(SharpUNet, self).__init__() self.enc1 = self.conv_block(3, 64) self.enc2 = self.conv_block(64, 128) self.enc3 = self.conv_block(128, 256) self.bottleneck = self.conv_block(256, 512) self.up1 = nn.Upsample(scale_factor=2, mode='bilinear', align_corners=True) self.dec1 = self.conv_block(512 + 256, 256) self.up2 = nn.Upsample(scale_factor=2, mode='bilinear', align_corners=True) self.dec2 = self.conv_block(256 + 128, 128) self.up3 = nn.Upsample(scale_factor=2, mode='bilinear', align_corners=True) self.dec3 = self.conv_block(128 + 64, 64) self.final = nn.Conv2d(64, 3, kernel_size=1) def conv_block(self, in_c, out_c): return nn.Sequential( nn.Conv2d(in_c, out_c, 3, padding=1), nn.BatchNorm2d(out_c), nn.ReLU(inplace=True), nn.Conv2d(out_c, out_c, 3, padding=1), nn.BatchNorm2d(out_c), nn.ReLU(inplace=True) ) def forward(self, x): e1 = self.enc1(x) e2 = self.enc2(F.max_pool2d(e1, 2)) e3 = self.enc3(F.max_pool2d(e2, 2)) b = self.bottleneck(F.max_pool2d(e3, 2)) d1 = self.up1(b) if d1.size() != e3.size(): d1 = F.interpolate(d1, size=e3.shape[2:]) d1 = torch.cat((d1, e3), dim=1) d1 = self.dec1(d1) d2 = self.up2(d1) if d2.size() != e2.size(): d2 = F.interpolate(d2, size=e2.shape[2:]) d2 = torch.cat((d2, e2), dim=1) d2 = self.dec2(d2) d3 = self.up3(d2) if d3.size() != e1.size(): d3 = F.interpolate(d3, size=e1.shape[2:]) d3 = torch.cat((d3, e1), dim=1) d3 = self.dec3(d3) return torch.sigmoid(self.final(d3)) # ========================= # 3. LOAD MODEL # ========================= model = SharpUNet().to(DEVICE) if not os.path.exists(MODEL_PATH): raise FileNotFoundError(f"Model not found at {MODEL_PATH}") model.load_state_dict(torch.load(MODEL_PATH, map_location=DEVICE)) model.eval() print("Model loaded successfully.") # ========================= # 4. INFERENCE FUNCTION # ========================= transform = transforms.Compose([ transforms.Resize((256, 256)), transforms.ToTensor() ]) def infer_image(user_img): # Convert uploaded image to RGB PIL img = user_img.convert("RGB") # Preprocess img_tensor = transform(img).unsqueeze(0).to(DEVICE) with torch.no_grad(): pred = model(img_tensor) pred_np = pred.squeeze(0).cpu().permute(1, 2, 0).numpy() pred_clean = pred_np.copy() pred_clean[pred_clean < 0.1] = 0 # Remove weak pixels for sharp display # Resize original image to match output size img_resized = img.resize((256, 256)) return img_resized, Image.fromarray((pred_clean * 255).astype(np.uint8)) # ========================= # 5. GRADIO INTERFACE # ========================= iface = gr.Interface( fn=infer_image, inputs=gr.Image(type="pil"), outputs=["image", "image"], title="""Automated Pseudo-Color Edge Detection in Images Using U-Net""", description="Upload any RGB image and see the predicted pseudo-colored edges." ) if __name__ == "__main__": iface.launch()