search

yhhhli / BRECQ

Pytorch implementation of BRECQ, ICLR 2021
MIT License
247 stars 56 forks source link

how to reproduce zero data result? #6

Closed yyfcc17 closed 3 years ago

yyfcc17 commented 3 years ago

as title.

there is a bug: https://github.com/yhhhli/BRECQ/blob/da93abc4f7e3ef437b356a2df8a5ecd8c326556e/main_imagenet.py#L173

args.batchsize should be args.workers

yhhhli commented 3 years ago

Thanks,

yyfcc17 commented 3 years ago

question not answered, how can i reproduce zero data result? --num_samples 0 , will triger an error

yhhhli commented 3 years ago

Do you mean the ZeroQ dataset? I use the code from https://github.com/amirgholami/ZeroQ, you can use that code to generate 1024 synthesized images

yyfcc17 commented 3 years ago

in your paper, there are results of BRECQ without training data: image

yhhhli commented 3 years ago

As I said, I use ZeroQ to generate 1024 synthesized images and then use them to perform BRECQ. You can find it in Section B.2 image image

yyfcc17 commented 3 years ago

oh, i see. thank you!