dhanajitb / petal Goto Github PK

Hello,
Thank you for your good work.
Could you please provide the link to ./ckpt/imagenet/corruptions/Standard_R50_swa.pt and ./ckpt/imagenet/corruptions/Standard_R50_cov.pt? I tried to run the train-swag-diagonal-imagenet.py file, but the save operation cost too much time and seemed never to stop.
By the way, in the train-swag-diagonal-imagenet.py file, since the batch_idx always equals 0, the optimizer and the swa_scheduler seem never to work. Is it intended?
Looking forward to your replay.

Hello
Thank you for your good work.

I have a question about your DataLoader for ImageNet-C.
When implementing PETAL for ImageNet-C, I found that your loader returns only 5,000 samples for each corruption, not all 50,000 samples.

In your configuration file, you specify # of examples for simulating test streams. However, the dataloader only serves the specified samples of the RobustBench library (robustbench/data/imagenet_test_image.ids.txt).

Is it intended?
I appreciate any help you can provide.