Apply same transform to image and mask?

[tom-osika]

Hi All,

I have a question about applying random transforms. Say I have two composite transforms, one for training images (x values), and one for training masks (y values). Each of these composite transforms contain random transformations (like flipping, zooming, etc.). I need to make sure that any random transforms that are applied to the image are also applied to the mask. Is set_determinism the best way to do this? The https://github.com/Project-MONAI/tutorials/blob/master/2d_segmentation/torch/unet_training_array.py example doesn't call set_determinism.

Thank you!

[wyli]

set_determinism will set the global seed. you can also use set_random_state of Compose which is more convenient for your use case.

The ArrayDataset used in the demo script calls transform.set_random_state(seed=self._seed) to ensure the transforms are synchronised. At each getitem, self._seed is refreshed

MONAI/monai/data/dataset.py

Line 1080 in e223932

def __getitem__(self, index: int):

[tom-osika]

So ArrayDatasets will automatically be synchronized? That's what I'm using in my case.