Closed erak closed 3 years ago
Hey, I just found this: https://github.com/una-dinosauria/3d-pose-baseline for 3d reconstruction from 2d data i.e. 1 camera. This seems promising. It's probably also worth to check out the forks that build on it.
There also is a scientific paper attached to it which is nice, since that means there is some documentation/explanation attached to it
I will look into https://github.com/facebookresearch/VideoPose3D since I got the AMD GPU setup working with PyTorch.
Hey, I just found this: https://github.com/una-dinosauria/3d-pose-baseline for 3d reconstruction from 2d data i.e. 1 camera. This seems promising. It's probably also worth to check out the forks that build on it.
@nikozoe Nice, it only requires tensorflow > 1.0, and there is a special build of tensorflow 1.14 for rocm.
There is also the service I mentioned the other day: https://getrad.co. It's quite expensive though (taken a monthly payment for the Studio plan https://getrad.co/studio-product/ into account). On the other hand, they would provide FBX files, see #5.
Even on premise execution seem to have limitations on the rendered output.
Closing this now, since we now have a working system based on detectron2 and VideoPose3D.
https://github.com/CMU-Perceptual-Computing-Lab/openpose/