fix(mask): improve mask loading to match standard convention (white=remove); add debug logs

api/main.py

@@ -125,16 +125,30 @@ def _load_rgba_image(path: str) -> Image.Image:
 
 
 def _load_rgba_mask_from_image(img: Image.Image) -> np.ndarray:
-    # Expected by process_inpaint: RGBA where alpha=0 for drawn (to remove), 255 elsewhere
+    # Standard convention: white=remove (255), black=keep (0)
+    # Convert to RGBA where alpha=0 means "to remove", alpha=255 means "keep"
     if img.mode != "RGBA":
-        # If no alpha, treat non-black/white>0 as masked areas
+        # For RGB/Grayscale masks: white (value>128) = remove, black (value<=128) = keep
         gray = img.convert("L")
         arr = np.array(gray)
-        alpha = np.where(arr > 0, 0, 255).astype(np.uint8)
+        # White pixels (>128) should have alpha=0 (to remove), black pixels (<=128) alpha=255 (keep)
+        alpha = np.where(arr > 128, 0, 255).astype(np.uint8)
         rgba = np.zeros((img.height, img.width, 4), dtype=np.uint8)
         rgba[:, :, 3] = alpha
         return rgba
-    return np.array(img)
+    # For RGBA: check if alpha channel is used or RGB channels
+    arr = np.array(img)
+    alpha = arr[:, :, 3]
+    # If alpha is mostly opaque (mean > 200), treat RGB channels as mask values
+    if alpha.mean() > 200:
+        # Use RGB to determine mask: white in RGB = remove
+        gray = cv2.cvtColor(arr[:, :, :3], cv2.COLOR_RGB2GRAY)
+        alpha = np.where(gray > 128, 0, 255).astype(np.uint8)
+        rgba = arr.copy()
+        rgba[:, :, 3] = alpha
+        return rgba
+    # Alpha channel already encodes the mask
+    return arr
 
 
 @app.post("/inpaint")

src/core.py

@@ -459,10 +459,16 @@ def process_inpaint(image, mask, invert_mask=True):
     image = norm_img(image)
 
     # Convert RGBA mask to single-channel mask.
-    # When invert_mask=True (default), areas the user paints (alpha=0) become 255 (to remove).
+    # Standard: white=remove (255), black=keep (0)
+    # When invert_mask=True (default): alpha=0 (transparent/painted) → 255 (remove), alpha=255 → 0 (keep)
     alpha_channel = mask[:,:,3]
     mask = (255 - alpha_channel) if invert_mask else alpha_channel
     mask = resize_max_size(mask, size_limit=size_limit, interpolation=interpolation)
+    
+    # Debug: log mask statistics
+    mask_nonzero = int((mask > 128).sum())
+    print(f"Mask shape: {mask.shape}, non-zero pixels (>128): {mask_nonzero}")
+    
     mask = norm_img(mask)
 
     res_np_img = run(image, mask)