Qing Shuai

It seems that the optimization falls into local optimal and generates a bad result.

Please use new pipeline now: ```bash export openpose= python3 apps/preprocess/extract_keypoints.py ${data} --mode openpose --hand --face ```

Maybe there is some inconsistent in the model config of hand. In the past maybe I used `use_flat_mean=True`. I will check it later.

Hello, you can print the filename when the error occurs.

Seems the file is corrupted. You can remove the `annots` folder and generate again.

No. This won't cause the decode error.

Thanks @NishilVani . Maybe it's a bug in windows? I didn't test the code on Windows now.

This repo's implementation is not as the same as the origin. This [operation](https://github.com/zju3dv/EasyMocap/blob/master/doc/mvmp.md#1-reconstucting-human-pose) have the similar functions for this task.

> there are many wrong in demo with openpose, so do you have a plan to fix it ? What's wrong do you mean? > and i notice the args...

Hello, thanks for your attention. I implement this by set the vposer variables as the optimization parameters. When calculate the loss, I first decode the `poses` and forward the SMPL...