Bug fixes in read_image_to_tensor (#97)

* Select cpu if gpu not available * Change BGR to RGB before inference * Append sys path, so the notebook can open on windows * Convert color from bgr to rgb * Remove sys path because we assume that the notebooks are decoupled with the path dependence * Remove color conversion because it perform in the function read_image_to_tensor
zhiqwang · Apr 21, 2021 · f4bb74c · f4bb74c
1 parent dbe2d4d
commit f4bb74c
Show file tree

Hide file tree

Showing 2 changed files with 34 additions and 32 deletions.
diff --git a/notebooks/inference-pytorch-export-libtorch.ipynb b/notebooks/inference-pytorch-export-libtorch.ipynb
diff --git a/yolort/utils/image_utils.py b/yolort/utils/image_utils.py
@@ -175,6 +175,7 @@ def read_image_to_tensor(
         image (np.ndarray): the candidate ndarray image to be parsed to Tensor.
         is_half (bool): whether to transfer image to half. Default: False.
     """
+    image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
     image = np.ascontiguousarray(image, dtype=np.float32)  # uint8 to float32
     image = np.transpose(image / 255.0, [2, 0, 1])