HandOccNet Reproduce DexYCB results

Hi,

Thank you for producing such fantastic work for the community. I followed the training instructions and trained a model of DexYCB for 25 epochs. I got very poor results:

MPJPE : 55.41 mm PA MPJPE : 14.15 mm

Can the authors please shed light on where I may have gone astray?

Thank you very much!

Sep 28 '22 07:09 kfzyqin

Could you capture your train_logs.txt (saved in output/log) and share it ?

And did you get the same results with ours by using pretrained model ?

Sep 28 '22 07:09 namepllet

The pretrained model's results are much better:

MPJPE: 10.01 mm
PA MPJPE: 4.52 mm You can get my train_logs.txt through this link: https://drive.google.com/file/d/1c_2qpufOuWGM_KmljYQpiXOm2em2iDZx/view?usp=sharing

Sep 28 '22 07:09 kfzyqin

It seems you used GPUs less than 4.

We used 4 GPUs, so you should scale learning rate according to your batch size.

Please refer https://github.com/namepllet/HandOccNet/issues/9#issuecomment-1217812789

Sep 28 '22 08:09 namepllet

Thank you :-) Let me give a try.

Sep 28 '22 08:09 kfzyqin

Thank you :-) Let me give a try.

Hi, I also have the same issue. Did you achieve better results after adjusting the learning rate and batch size?

Sep 05 '23 07:09 LiaoQi98