Sensitivity to pre-trained net
Thank you for your high performing submission.
I see that you use a pre-trained net in https://github.com/crim-ca/crims2s/blob/master/crims2s/training/model/conv_pretrained.py.
- Would your overall approach work without it?
- How sensitive is the RPSS to using such a pretrained net? (Does maybe one of your earlier tags just miss that pre-trained step? - no need to recompute just for this question out of curiosity)