Color Augmentation before of after Normalisation for ResNet

vision
1

I want to apply color augmentation by applying a transfrom.ColorJitter to my dataset that I feed into a Resnet. I’m not sure though whether I should normalise the image using the ImageNet mean and STD before or after the augmentation.

Any ideas what’s the correct approach?

Many thanks

2

In the samples I have seen so far, the ColorJitter is used prior to normalizing.
You can check this out : fastai_imagenet training script

    train_tfms = transforms.Compose([
        transforms.RandomResizedCrop(size),
        transforms.RandomHorizontalFlip(),
        transforms.ColorJitter(.4,.4,.4),
        transforms.ToTensor(),
        Lighting(0.1, __imagenet_pca['eigval'], __imagenet_pca['eigvec']),
        normalize,
    ])