zzzace2000 / robust_cls_model Goto Github PK

The code to reproduce CVPR 2021 paper "Towards Robust Classification Model by Counterfactual and Invariant Data Generation"

License: Apache License 2.0

Python 98.17% Shell 1.83%

robust_cls_model's People

Contributors

Stargazers

Watchers

Forkers

chnxindong

robust_cls_model's Issues

What does the cf argument mean?

robust_cls_model/train_pl.py

Line 56 in c2b9a79

 parser.add_argument("--cf", type=str, default='logp') # ['logp', 'uni', 'uni_e', 'channels'] 

robust_cls_model/arch/lightning/base.py

Line 406 in 9f52689

def counterfactual_ce_loss(self, logit, y, reduction='none'):

I've noticed that the cf value would change the calculation of the counterfactual loss, but I don't really understand that.

arch.lightning has no module named base_imaginet

When I re-implemented with this code, I found the following error :

Traceback (most recent call last):
  File "train_pl.py", line 18, in <module>
    from arch.lightning.csv_recording import CSVRecordingCallback
  File "/home/{$username}/robust_cls_model/arch/lightning/csv_recording.py", line 12, in <module>
    from .base_imagenet import ImageNetLightningModel
ModuleNotFoundError: No module named 'arch.lightning.base_imagenet'

I think this branch missed about base_imagenet.py in the directory of "/arch/lightning". or csv_recording.py seems to contain the problem.

Recommend Projects