• regression
• GAN
• Vnect

Now I only use h36m as a datasets and have to use mpii_inf_3dhp.

This is the simplest way of estimation something. it have already implemented. but showing result is too bad. the camera calibration and so on are tiresome.

This is my idea of estimation bone position. perhaps bone position is generated by distribution in 3d space

Vnect is one of the famous way to estimate pose. See the detail on it. http://gvv.mpi-inf.mpg.de/projects/VNect/

Train:
python main.py

Test:
python demo.py

See the resultimages directory.