fix

app.py

@@ -99,7 +99,7 @@ with block:
         cv2.circle(prompt_image, (x, y), 5, color, -1)
         point_coords.append([x,y])
         point_labels.append(point_label_radio)
-        sam_masks = sam_cpu.cond_pred(pts=np.array(point_coords), lbls=np.array(point_labels), embedding=embedding)
+        sam_masks = sam_cpu.cond_pred(pts=np.array(point_coords), lbls=np.array(point_labels), embedding=embedding, image_size=input_image.size)
         return  [ prompt_image,
                   (input_image, sam_masks),
                   point_coords,

inference.py

@@ -232,6 +232,7 @@ class CustomSamPredictor(SamPredictor):
 
     def decode_and_predict(
         self,
+        image_size: Tuple[int, int],
         embedding: torch.Tensor,
         point_coords: Optional[np.ndarray] = None,
         point_labels: Optional[np.ndarray] = None,
@@ -252,7 +253,7 @@ class CustomSamPredictor(SamPredictor):
           (np.ndarray): An array of quality predictions for each mask.
           (np.ndarray): Low resolution mask logits for subsequent iterations.
         """
-        self.set_torch_image(embedding, (embedding.shape[-2], embedding.shape[-1]))
+        self.features = embedding
         return self.predict(
             point_coords=point_coords,
             point_labels=point_labels,