How feasible it is to use ShakeDrop regularization in torchvision models? Anyone has tried this before?
torchvision