two places using multi-scale tricks

[Libaishun]

I see you use multi-scale tricks in two places: directly resizing model inputs in train.py and scale image with random_affine in datasets.py, which one is better to use or shoule we use them simultaneously ？

[glenn-jocher]

@Libaishun yes you are correct! Both methods are suitable for performing multi-scale image augmentation. Exact results depend on model, dataset, etc.

[Libaishun]

@glenn-jocher setting multi-scale=True in train.py seems use much more cuda memory than disable it.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.