[ONNX] fix dynamic resize (#2488)

* Update transform.py

* Update transform.py

Author: neginraoof

torchvision/models/detection/transform.py

@@ -20,7 +20,7 @@ def _resize_image_and_masks_onnx(image, self_min_size, self_max_size, target):
     scale_factor = torch.min(self_min_size / min_size, self_max_size / max_size)
 
     image = torch.nn.functional.interpolate(
-        image[None], scale_factor=scale_factor, mode='bilinear',
+        image[None], scale_factor=scale_factor, mode='bilinear', recompute_scale_factor=True,
         align_corners=False)[0]
 
     if target is None:
@@ -42,7 +42,7 @@ def _resize_image_and_masks(image, self_min_size, self_max_size, target):
     if max_size * scale_factor > self_max_size:
         scale_factor = self_max_size / max_size
     image = torch.nn.functional.interpolate(
-        image[None], scale_factor=scale_factor, mode='bilinear',
+        image[None], scale_factor=scale_factor, mode='bilinear', recompute_scale_factor=True,
         align_corners=False)[0]
 
     if target is None: