the source code of thesis "3d Pose Estimation with Joint Group Spatial-Temporal Transformer".

This neural network takes 2d pose sequence as input and output corresponding 3d pose sequence.

These are the predicted video output:

Acknowledgement: Some source codes of this project are from MHFormer, and thanks to Their excellent work.

Thank 3D Artist Jian Yuan for providing the reference videos.

I am improving this project.