Can you provide pretrained weights?

[JinYong Kim]

Thanks for your implementation.
Can you provide the pretrained weights for this model?

[L]

Can you provide the model_best.pth.tar, thanks @ziniuwan

[ZiNiU WaN]

Hi! I have added the google drive link of a Pytorch checkpoint file in the README file. Good luck!

[L]

thanks @ziniuwan , I found that I used this model, and the results were not consistent with your report，such as “PVE”, so on

CUDA_VISIBLE_DEVICES=0 python eval.py --cfg configs/config_stage2.yaml --pretrained ./model_data/model_best.pth.tar --eval_set test

[hdjang]

Hi @zhLawliet

When I use the same command line as you used, I got the following errors. How did you solve this?

[L]

@hdjang change the view by reshape
rel_joints.reshape(-1, 3, 1).reshape(-1, joints.shape[1], 4, 4)

[ZiNiU WaN]

@hdjang This is a version conflict between smplx and PyTorch. We also found the errors during experiments. Modifying all the view() to reshape() wherever the errors indicate in the smplx code as @zhLawliet said could solve the problems.

[hdjang]

Thanks for instant replies! @zhLawliet , @ziniuwan