JiahuiYu/slimmable_networks

model profiling

lyn0102 opened this issue · 3 comments

lyn0102 commented

Hello! I compared the results of statistical FLOPs and parameters in your code with the results in the thop tool, and found that there are some differences. I guess the reason is that you did not count the FLOPs and parameters of the bn layer? Looking forward to your answer, thank you!

JiahuiYu commented

I think all papers do not include BN as they can be fused during inference.

  • 👍1
lyn0102 commented

Thank you for your reply :D

abhishekaich27 commented

Hi @JiahuiYu, are you calculating MACs or Flops? The code says "n_macs".

Asking to be sure and make necessary changes.