Update utils_mask.py

utils_mask.py

@@ -78,25 +78,28 @@ def get_mask_location(model_type, category, model_parse: Image.Image, keypoint:
     arms_right = (parse_array == 15).astype(np.float32)
 
     if category == 'dresses':
-        parse_mask = (parse_array == 7).astype(np.float32) + \
-                    (parse_array == 4).astype(np.float32) + \
-                    (parse_array == 5).astype(np.float32) + \
-                    (parse_array == 6).astype(np.float32)
+        # Initial dress mask without legs
+        parse_mask_upper = (parse_array == 7).astype(np.float32) + \
+                        (parse_array == 4).astype(np.float32) + \
+                        (parse_array == 5).astype(np.float32)
 
-        parser_mask_changeable += np.logical_and(parse_array, np.logical_not(parser_mask_fixed))
+        # Create a separate mask for the legs
+        parse_mask_legs = (parse_array == 12).astype(np.float32) + \
+                        (parse_array == 13).astype(np.float32) + \
+                        (parse_array == 6).astype(np.float32)
 
-        # Create a separate mask for the leg parts
-        leg_mask = (parse_array == 12).astype(np.float32) + \
-                (parse_array == 13).astype(np.float32)
+        parser_mask_changeable += np.logical_and(parse_array, np.logical_not(parser_mask_fixed))
 
         # Fill gaps between legs
-        leg_gap_filled = cv2.dilate(leg_mask.astype(np.uint8), np.ones((10, 10), np.uint8), iterations=10)
+        leg_mask_filled = cv2.dilate(parse_mask_legs.astype(np.uint8), np.ones((6, 6), np.uint8), iterations=6)
+        leg_mask_filled = cv2.erode(leg_mask_filled, np.ones((6, 6), np.uint8), iterations=6)  # Optional: Erode back to original size
+
+        # Combine filled leg mask with upper body mask
+        parse_mask = np.maximum(parse_mask_upper, leg_mask_filled)
 
-        # Ensure the filled leg gaps do not affect the upper body
-        leg_gap_filled = np.logical_and(leg_gap_filled, leg_mask).astype(np.float32)
+        # Final dilation to ensure there are no small gaps
+        parse_mask = cv2.dilate(parse_mask.astype(np.uint8), np.ones((5, 5), np.uint8), iterations=5)
 
-        # Add the filled leg gaps to the parse_mask
-        parse_mask += leg_gap_filled
 
 
     elif category == 'upper_body':