search

namepllet / HandOccNet

Offical pytorch implementation of "HandOccNet: Occlusion-Robust 3D Hand Mesh Estimation Network", CVPR 2022.
155 stars 30 forks source link

Reproduce DexYCB results #16

Open kfzyqin opened 1 year ago

kfzyqin commented 1 year ago

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!

namepllet commented 1 year ago

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 ?

kfzyqin commented 1 year ago

The pretrained model's results are much better:

namepllet commented 1 year ago

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

kfzyqin commented 1 year ago

Thank you :-) Let me give a try.

LiaoQi98 commented 1 year ago

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?