This module is for converting 2d human upper body poses to 3d. It is a part of co-speech gesture generation project. You can find out more details in the paper.
- download CMU panoptic dataset
- run generate_dataset.py to make a single pickle file for training
- run train.py to train the simple neural network that estimates depth values from 2d poses
- run test.py to test the trained model