The effect of the `base_acc` in accuracy_predictor ?

mit-han-lab / once-for-all

[ICLR 2020] Once for All: Train One Network and Specialize it for Efficient Deployment

https://ofa.mit.edu/

MIT License

1.89k stars 333 forks source link

The effect of the `base_acc` in accuracy_predictor ? #42

Open lucienne999 opened 3 years ago

lucienne999 commented 3 years ago

Hi, I noticed that here(link1) add base acc to train accuracy predictor and set the last linear layer bias as False. For the same batch data, they have the same bias. If I do not add this bias, is the predictor will decrease? Or I missing any analysis in the paper?

link1 : https://github.com/mit-han-lab/once-for-all/blob/master/ofa/nas/accuracy_predictor/acc_predictor.py#L32

kourner commented 2 years ago

I have no idea how to train the accuracy predictor, can you share your training script to me? Thanks a million.

Fr1ck commented 2 years ago

I followed the author's explanation at #30 and replaced the zero tensor with average accuracy. However, the predictor still can not converge well T_T

kourner commented 2 years ago

I followed the author's explanation at #30 and replaced the zero tensor with average accuracy. However, the predictor still can not converge well T_T

我打算做一个ofa的detection版本，但目前卡在训练acc_pridictor了。如果你也感兴趣的话，可以加我一起讨论。这是我的联系方式QQ/微信：953883