Reproducing ZSL mAP with tresnet-m lr 10e-4 and wd 10e-2

Alibaba-MIIL / ML_Decoder

Official PyTorch implementation of "ML-Decoder: Scalable and Versatile Classification Head" (2021)

MIT License

317 stars 53 forks source link

Reproducing ZSL mAP with tresnet-m lr 10e-4 and wd 10e-2 #33

Closed aliman80 closed 2 years ago

aliman80 commented 2 years ago

Hi, I am using ZSL code with tresnetn m lr 10e-4 and wdecay 10e-2 , my mAP after 10 epochs were 18.94 and now its 27th epoch and it started decreasing instead of increasing so the highest till now is 18.94. Can you guide me what am i missing here. Will be great help for me thanks again.

giladsharir commented 2 years ago

Hi , We fixed an issue in the ZSL training code.
You should now get higher scores. (with the default learning rate and weight-decay).

aliman80 commented 2 years ago

Thank you so much.Regards