EverybodyDanceNow_reproduce_pytorch
EverybodyDanceNow_reproduce_pytorch copied to clipboard
What are challenges for hand refinement?
I am currently using Pix2PixHD as generator in my Master's thesis. I wanted to know from Author's perspective the challenges you are facing in achieving Hand refinement.
I'm not familiar enough with Pix2PixHD but a new Hand-gesture detector by google researchers hit the shelf after is showcase on CVPR 2019. I suppose a step in the pipe line would be to detect and crop out the hand pixels no?
I was tinkering with the model and trying to make it run (for hand tracking only, and outside of tfliite). I shared my findings in this notebook.
What else would be needed?
EDIT: Looking at the original paper seems like keypoints of fingers are expected from the pose estimator model. Not sure why these are not available in this implementation.