Update app.py

app.py

@@ -4,51 +4,18 @@ import torch
 import cv2
 from PIL import Image
 from torchvision import transforms
-from cloth_segmentation.networks.u2net import U2NET  # Import U²-Net
+from cloth_segmentation.networks.u2net import U2NET
 
-# Load U²-Net model
+# Load U²-Net Model
 model_path = "cloth_segmentation/networks/u2net.pth"
 model = U2NET(3, 1)
 state_dict = torch.load(model_path, map_location=torch.device('cpu'))
-state_dict = {k.replace('module.', ''): v for k, v in state_dict.items()}  # Remove 'module.' prefix
+state_dict = {k.replace('module.', ''): v for k, v in state_dict.items()}
 model.load_state_dict(state_dict)
 model.eval()
 
-def detect_design(image_np):
-    """Detects if a design exists on the dress using edge detection & clustering."""
-    gray = cv2.cvtColor(image_np, cv2.COLOR_RGB2GRAY)
-    edges = cv2.Canny(gray, 50, 150)
-
-    # Dilation to highlight patterns
-    kernel = np.ones((3, 3), np.uint8)
-    edges = cv2.dilate(edges, kernel, iterations=1)
-
-    # Count edge density
-    design_ratio = np.sum(edges > 0) / (image_np.shape[0] * image_np.shape[1])
-
-    return design_ratio > 0.02, edges  # If edge density is high, assume a design is present
-
 def segment_dress(image_np):
-    """Segment the dress using U²-Net & refine with Lab color space."""
-    
-    # Convert to Lab space
-    img_lab = cv2.cvtColor(image_np, cv2.COLOR_RGB2LAB)
-    L, A, B = cv2.split(img_lab)
-
-    # Use K-means clustering to detect dominant dress region
-    pixel_values = img_lab.reshape((-1, 3)).astype(np.float32)
-    k = 3  # Three clusters: background, skin, dress
-    _, labels, centers = cv2.kmeans(pixel_values, k, None, (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 10, 1.0), 10, cv2.KMEANS_RANDOM_CENTERS)
-    labels = labels.reshape(image_np.shape[:2])
-    
-    # Assume dress is the largest non-background cluster
-    unique_labels, counts = np.unique(labels, return_counts=True)
-    dress_label = unique_labels[np.argmax(counts[1:]) + 1]  # Avoid background
-    
-    # Create dress mask
-    mask = (labels == dress_label).astype(np.uint8) * 255
-
-    # Use U²-Net prediction to refine segmentation
+    """Segment the dress using U²-Net and GrabCut."""
     transform_pipeline = transforms.Compose([
         transforms.ToTensor(),
         transforms.Resize((320, 320))
@@ -59,49 +26,48 @@ def segment_dress(image_np):
 
     with torch.no_grad():
         output = model(input_tensor)[0][0].squeeze().cpu().numpy()
-
+    
     u2net_mask = (output > 0.5).astype(np.uint8) * 255
     u2net_mask = cv2.resize(u2net_mask, (image_np.shape[1], image_np.shape[0]), interpolation=cv2.INTER_NEAREST)
-    
-    # Combine K-means and U²-Net masks
-    refined_mask = cv2.bitwise_and(mask, u2net_mask)
 
-    # Morphological operations for smoothness
-    kernel = np.ones((5, 5), np.uint8)
-    refined_mask = cv2.morphologyEx(refined_mask, cv2.MORPH_CLOSE, kernel)
-    refined_mask = cv2.GaussianBlur(refined_mask, (15, 15), 5)
+    # Apply GrabCut to refine the mask
+    mask = np.zeros(image_np.shape[:2], np.uint8)
+    mask[u2net_mask > 128] = cv2.GC_FGD
+    mask[u2net_mask <= 128] = cv2.GC_BGD
+    bg_model = np.zeros((1, 65), np.float64)
+    fg_model = np.zeros((1, 65), np.float64)
     
-    return refined_mask
+    cv2.grabCut(image_np, mask, None, bg_model, fg_model, 5, cv2.GC_INIT_WITH_MASK)
+    mask = np.where((mask == 2) | (mask == 0), 0, 255).astype(np.uint8)
+    
+    return mask
 
-def recolor_dress(image_np, mask, target_color, edges):
-    """Change dress color while preserving texture, shadows, and designs."""
+def recolor_dress(image_np, mask, target_color):
+    """Recolor the dress while keeping texture, shadows, and designs."""
     
+    # Convert to LAB color space
     img_lab = cv2.cvtColor(image_np, cv2.COLOR_RGB2LAB)
+    
+    # Target color in LAB
     target_color_lab = cv2.cvtColor(np.uint8([[target_color]]), cv2.COLOR_BGR2LAB)[0][0]
     
-    # Exclude design from recoloring
-    design_mask = (edges > 0).astype(np.uint8) * 255
-    mask = cv2.bitwise_and(mask, cv2.bitwise_not(design_mask))
-
     # Preserve lightness (L) and change only chromatic channels (A & B)
-    blend_factor = 0.7
+    blend_factor = 0.8
     img_lab[..., 1] = np.where(mask > 128, img_lab[..., 1] * (1 - blend_factor) + target_color_lab[1] * blend_factor, img_lab[..., 1])
     img_lab[..., 2] = np.where(mask > 128, img_lab[..., 2] * (1 - blend_factor) + target_color_lab[2] * blend_factor, img_lab[..., 2])
 
+    # Convert back to RGB
     img_recolored = cv2.cvtColor(img_lab, cv2.COLOR_LAB2RGB)
     return img_recolored
 
 def change_dress_color(image_path, color):
-    """Change the dress color naturally while keeping designs intact."""
+    """Change the dress color while preserving texture and design details."""
     if image_path is None:
         return None
 
     img = Image.open(image_path).convert("RGB")
     img_np = np.array(img)
 
-    # Detect if a design is present
-    design_present, edges = detect_design(img_np)
-
     # Get dress segmentation mask
     mask = segment_dress(img_np)
     
@@ -117,12 +83,7 @@ def change_dress_color(image_path, color):
     new_color_bgr = np.array(color_map.get(color, (0, 0, 255)), dtype=np.uint8)  # Default to Red
     
     # Apply recoloring logic
-    if design_present:
-        print("Design detected! Coloring only non-design areas.")
-        img_recolored = recolor_dress(img_np, mask, new_color_bgr, edges)
-    else:
-        print("No design detected. Coloring entire dress.")
-        img_recolored = recolor_dress(img_np, mask, new_color_bgr, np.zeros_like(mask))  # No design mask
+    img_recolored = recolor_dress(img_np, mask, new_color_bgr)
 
     return Image.fromarray(img_recolored)