Can you share the found policy weights for the image classification benchmarks?

[saigontrade88]

Hi,

I am trying to replicate your work. I can't find your found policies's weights on these benchmarks yet. I looked into Fig.5 in the Faster AutoAugment paper. For confirmation, does the Figure show the complete found policies and their "weights"? In the Fig.5, are the found policies based on the full CIFAR-10 training set or the subset?

I also looked into [the train.py] program at . The code shows
path = Path(hydra.utils.get_original_cwd()) / cfg.path
What does Path(hydra.utils.get_original_cwd())return? I can't find the path cfg.path to the found policies yet.

Can you help point me to the files with such information? For example, in the Fast AA, readers can find their policies at this link https://github.com/kakaobrain/fast-autoaugment/blob/master/FastAutoAugment/archive.py

Any comments is appreciated. Thanks.

[moskomule]

No, it only shows some selected subpolicies.

I implemented this codebase long before, so I don't remember the codes.
AutoAlbument https://albumentations.ai/docs/autoalbument/ is a re-implementation of our work with better documentation, so I hope it will help you.

[saigontrade88]

Many thanks for your prompt reply. I appreciate your input and will look into the Autoalbument for their found policies and weights. You also suggest looking at the computer vision library in #9. However, there is a severe complaint about the search process being slow at this link. Can you share your experience with the autoalbument library? Many thanks.

[moskomule]

Unfortunately, I have never used autoalbument.