WeakNAS

WeakNAS copied to clipboard

Hello, great work! I wonder how to get the pre-processed xxx Subset results by myself on other benchmarks? It looks like the result is obtained by running OFA. Could you please provide simple instructions about how to obtain the results? I think it is very beneficial for the community because we can figure out the whole pipeline.

BTW, another question is , when I run the code, it will always go to AssertionError. There might be small typos in the code? WeakNAS-master/WeakNAS.py", line 384, in main assert args.train_set in setname_list AssertionError

Hi detectRecog,

Thanks for your interest in our paper! I just fix the AssertionError bug in the latest commits, can you try it again?

Regarding pre-processed subset: For the NASBench subset, We extract the result from their respective NASBench package. For ImageNet subset, We ran it through inference of OFA model, the current ImageNet experiments should be already runnable through imagenet_mobilenet_search.ipynb, It will do the inference inside the jupyter notebook. I also plan to release the inference accuracy of OFA model (to make it a bench) to accelerate the search process.

Best, Junru