Chapter 3: Image Augmentation

[imteekay]

There's this code for the image augmentation

# Transforms for the training data
train_transforms = transforms.Compose([
    transforms.RandomHorizontalFlip(),
    transforms.RandomRotation(20),
    transforms.RandomResizedCrop(150),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])  
])
 
# Transforms for the validation data
val_transforms = transforms.Compose([
    transforms.Resize(150),
    transforms.CenterCrop(150),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
])

I was wondering if we should only pre-process the validation data, or in this case, it's ok to also "CenterCrop" the val data?
I'm asking this question because I read that for training data, we can pre-process (resize, to-tensor, normalize) and augment (flip, rotate, crop), but for validation data, we should only pre-process, but not augment. Is this idea true?

If that's true, I should remove the CenterCrop from the validation transform, right?

val_transforms = transforms.Compose([
    transforms.Resize(150),
-   transforms.CenterCrop(150),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
])