I am not sure what the point of the data augmentations in PyTorch when loading images. Specifically, I am talking about this:
transform = transforms.Compose([ transforms.ToTensor(), A.Normalize(CIFAR10_MEAN, CIFAR10_STD_DEV), A.RandomCrop(height=128, width=128),])
and loading in the training dataset using
train = datasets.CIFAR10(root=path, train=True, download=True, transform=transform)
I checked the size of
train when I do no transformation, and when I do a lot of random transformations(such as random crop, random horizontal flip, etc.) and it seems like train is not getting larger(I checked using both
train is the same size no matter what transforms I do), so I was wondering what the point was.
I don't see how you can get better results if you don't actually append the data once we do these transformations to the dataset.
Any clarification or explanation would be greatly appreciated.