search

zhangy76 / PhysMoP

Code repository for Incorporating Physics Principles for Precise Human Motion Prediction
MIT License
22 stars 3 forks source link

The processing code of Human3.6M and 3DPW #5

Closed HitBadTrap closed 3 weeks ago

HitBadTrap commented 1 month ago

Great Work! However, I'm confused about which file I should download for Human3.6M from https://github.com/mks0601/NeuralAnnot_RELEASE. There are three links (data, SMPL parameters, and SMPL-X parameters).

Another issue is when will you share the processing code of Human3.6M and 3DPW? It's essential to me.

Thanks a lot.

zhangy76 commented 1 month ago

Hi, thanks, for Human3.6M data, please download https://1drv.ms/f/s!All7gdNh7XE5oRb-ZscrxlS75Ogb?e=EaFfQz.

For the preprocessing code of Human3.6M and 3DPW, as long as the data files are downloaded, you will find they share very similar format, you can directly use the same preprocessing code but only need change the folder path.

HitBadTrap commented 1 month ago

I notice that the data of H36m does not contain gender_id and framerate, what should I set them to? Moreover, why are the shape parameters different for each frame in H36m? And how many beta parameters you adopt for 3DPW?

zhangy76 commented 1 month ago

gender_id is 0, and framerate is 50 (downsample to 25).

the shape parameters are not perfect, we do an average.

10 beta parameters.

zhangy76 commented 3 weeks ago

Close as being inactive. Please feel free to open again, thanks.