What data augmentation strategies are used, when the pre-trained CNN models(eg. resnet50) in 'torchvision.models' are trained?

1

I tested the pre-trained resnet50 model in ‘torchvision.models’ on adjusted ImageNet validation-set whose brightness was decreased by a factor of 0.1. And an obvious accuracy decrease was found. So, I want to know what data augmentation strategies are used when the pre-trained CNN models(eg. resnet50) in ‘torchvision.models’ are trained?
Anyone could help me? thank you!

1 Like