Update app.py

removed the something new it makes the images worse

app.py

@@ -117,60 +117,18 @@ net = build_model(hypar, device)
 
 def refine_mask(mask):
     """
-    Refine the mask using advanced image processing techniques.
+    Softly refine the mask using Gaussian Blur and feathering for smooth transitions.
     """
-    # Convert to numpy array if it's not already
-    if isinstance(mask, Image.Image):
-        mask = np.array(mask)
-
-    # Apply binary thresholding
-    _, binary_mask = cv2.threshold(mask, 128, 255, cv2.THRESH_BINARY)
-
-    # Use morphological operations to clean up the mask
-    kernel = np.ones((5,5), np.uint8)
-    cleaned_mask = cv2.morphologyEx(binary_mask, cv2.MORPH_CLOSE, kernel)
-    cleaned_mask = cv2.morphologyEx(cleaned_mask, cv2.MORPH_OPEN, kernel)
-
-    # Edge detection
-    edges = cv2.Canny(cleaned_mask, 100, 200)
-
-    # Dilate edges to create a transition area
-    dilated_edges = cv2.dilate(edges, kernel, iterations=2)
-
-    # Create a gradient in the transition area
-    distance = cv2.distanceTransform(255 - dilated_edges, cv2.DIST_L2, 5)
-    distance = cv2.normalize(distance, None, 0, 1, cv2.NORM_MINMAX)
-
-    # Combine the cleaned mask with the gradient transition
-    refined_mask = cleaned_mask.astype(float) / 255.0
-    refined_mask = np.maximum(refined_mask, distance)
-    refined_mask = (refined_mask * 255).astype(np.uint8)
-
-    # Final smoothing
-    refined_mask = cv2.GaussianBlur(refined_mask, (5, 5), 0)
+    # Apply Gaussian Blur to soften edges and make the mask more continuous
+    refined_mask = cv2.GaussianBlur(mask, (5, 5), 0)
+    
+    # Feather the edges for a smoother transition between foreground and background
+    feathered_mask = cv2.copyMakeBorder(refined_mask, 10, 10, 10, 10, cv2.BORDER_CONSTANT, value=[255])
+    feathered_mask = cv2.GaussianBlur(feathered_mask, (21, 21), 0)
+    refined_mask = feathered_mask[10:-10, 10:-10]  # Remove border
 
     return refined_mask
-    
-def remove_background(image_path):
-    # Load the image
-    image = Image.open(image_path).convert("RGBA")
-    
-    # Create a mask (this is where you'd use the DIS model)
-    # For demonstration, let's create a simple circular mask
-    mask = Image.new("L", image.size, 0)
-    draw = ImageDraw.Draw(mask)
-    draw.ellipse((0, 0) + image.size, fill=255)
-    
-    # Refine the mask
-    refined_mask = refine_mask(mask)
-    
-    # Convert refined mask back to PIL Image
-    refined_mask_pil = Image.fromarray(refined_mask)
-    
-    # Apply the refined mask
-    image.putalpha(refined_mask_pil)
-    
-    return image
+
 def inference(image):
     image_path = image