Fix the shift issue of the prediction

README.md

@@ -26,9 +26,9 @@ The model in Keras is built by the essay author, see [sergiomsilva/alpr-unconstr
     </tr>
     <tr>
         <td> Confidence </td>
-        <td> 0.9705 </td>
-        <td> 0.9828 </td>
-        <td> 0.9987 </td>
+        <td> 0.9841 </td>
+        <td> 0.9945 </td>
+        <td> 0.9979 </td>
     </tr>
 </table>
 

wpodnet/backend.py

@@ -45,6 +45,7 @@ class Predictor:
         [1., 1., 1., 1.]
     ])
     _scaling_const = 7.75
+    _stride = 16
 
     def __init__(self, wpodnet: WPODNet):
         self.wpodnet = wpodnet
@@ -55,11 +56,15 @@ def _resize_to_fixed_ratio(self, image: Image.Image) -> Image.Image:
 
         wh_ratio = max(h, w) / min(h, w)
         side = int(wh_ratio * 288)
-        bound_dim = min(side + side % 16, 608)
+        bound_dim = min(side + side % self._stride, 608)
 
         factor = bound_dim / min(h, w)
         reg_w, reg_h = int(w * factor), int(h * factor)
 
+        # Ensure the both width and height are the multiply of `self._stride`
+        reg_w += self._stride - reg_w % self._stride
+        reg_h += self._stride - reg_h % self._stride
+
         return image.resize((reg_w, reg_h))
 
     def _to_torch_image(self, image: Image.Image) -> torch.Tensor: