Images are rotated when training

[zhaodadu111]

hello,
I'm now trying to train a segmentation model for ultrasound images with dims (3,256,256). When I'm training the model, the input images are rotated 90 degree to the left. how can i rotate it back? Should i use Orientation()?

Thank you,
Zhaoda

[KumoLiu]

Hi @zhaodadu111, do you mean that you have checked the input and you noticed that his orientation had altered during training? In theory, training does not cause the data to change orientation. If you find your input images' orientation indeed changed, you can rotate your image using Rotate90. Or you can provide more information like your pre-processing process.

[wyli]

maybe you run into this issue #4862, you can also try Transpose

>>> from monai.transforms import Transpose
>>> x = torch.zeros((3, 100, 101))
>>> Transpose((0, 2, 1))(x).shape
(3, 101, 100)