How to approach TTA with albumentations?

[sakvaua]

Hi.
Is there a recommended approach to do test time augmentation? I would like to pick several random augmentations and apply them to test and validation sets for later stacking. I'm not sure how to proceed. How do I make sure that one set of augmentations is applied to the whole test set and then to validation set and not random augmentations to every image?
Thanks.

[BloodAxe]

It is not supported out of the box, but you may use a simple trick to fix random seed before applying augmentation at test time:

# TTA
random.seed(1234)
data = transform(**data)

[albu]

you can use p=1 transforms or always_apply transforms if it's something like horizontal flip or vertical flip (determined except probability)

[BloodAxe]

@sakvaua Hi, albumentations does not support this by design. However this is popular task, so I take a chance and pr my library with handy PyTorch extensions including TTA: https://github.com/BloodAxe/pytorch-toolbelt/blob/develop/pytorch_toolbelt/inference/tta.py