Since the blog is released half a year ago, it is for 3.0-dp-py3 release.
Currently, the latest docker is 3.0-py3. There is some change in augmentation part.
So, please follow NVIDIA TAO Documentation to modify the augmentation part of your training spec file.