Normally we will put the augmentation part in the def getitem(self, index) of class TensorData_scale(Dataset), so that for every epoch, a different kind of augmentation could be applied to [img1, img2, mask].
I found that you did the augmentation before def getitem(self, index) and saved it as a fixed numpy array. I assume possibly such a data loading way might cause out of memory problem and fixed augmentation for patch image in every epoch.
Normally we will put the augmentation part in the def getitem(self, index) of class TensorData_scale(Dataset), so that for every epoch, a different kind of augmentation could be applied to [img1, img2, mask].
I found that you did the augmentation before def getitem(self, index) and saved it as a fixed numpy array. I assume possibly such a data loading way might cause out of memory problem and fixed augmentation for patch image in every epoch.