Introduce dim_min and dim_max to regulate input image dimensions (#…

…15)

* Introduction of the parameters Dmin and Dmax, which regulate the input image dimensions into the model

* Rename Dmin to dim_min and Dmax to dim_max

wpodnet/backend.py

@@ -51,12 +51,12 @@ def __init__(self, wpodnet: WPODNet):
         self.wpodnet = wpodnet
         self.wpodnet.eval()
 
-    def _resize_to_fixed_ratio(self, image: Image.Image) -> Image.Image:
+    def _resize_to_fixed_ratio(self, image: Image.Image, dim_min: int, dim_max: int) -> Image.Image:
         h, w = image.height, image.width
 
         wh_ratio = max(h, w) / min(h, w)
-        side = int(wh_ratio * 288)
-        bound_dim = min(side + side % self._stride, 608)
+        side = int(wh_ratio * dim_min)
+        bound_dim = min(side + side % self._stride, dim_max)
 
         factor = bound_dim / min(h, w)
         reg_w, reg_h = int(w * factor), int(h * factor)
@@ -103,12 +103,12 @@ def _get_bounds(self, affines: np.ndarray, anchor_y: int, anchor_x: int, scaling
 
         return np.transpose(bounds)
 
-    def predict(self, image: Image.Image, scaling_ratio: float = 1.0) -> Prediction:
+    def predict(self, image: Image.Image, scaling_ratio: float = 1.0, dim_min: int = 288, dim_max: int = 608) -> Prediction:
         orig_h, orig_w = image.height, image.width
 
         # Resize the image to fixed ratio
         # This operation is convienence for setup the anchors
-        resized = self._resize_to_fixed_ratio(image)
+        resized = self._resize_to_fixed_ratio(image, dim_min=dim_min, dim_max=dim_max)
         resized = self._to_torch_image(resized)
         resized = resized.to(self.wpodnet.device)