VIBE

Why we included this project

VIBE is the official implementation of the CVPR 2020 paper of the same name, and it remains a solid reference point if you need 3D body pose and shape from ordinary video. The method walks through each frame of a clip and returns SMPL body-model parameters, so you get a full articulated body rather than 2D keypoints, which is what animation, biomechanics, sports tracking, and AR/VR work usually need. Its main practical edge over single-frame estimators is temporal: a recurrent structure carries information across frames, and that visibly reduces the jitter that shows up in real footage. The repository includes both demo and training code in PyTorch plus a Colab notebook, so you can try it on your own clips without assembling a pipeline from scratch. One caveat before you commit: the code and pretrained models ship under a non-commercial research license and depend on the SMPL body model, which has its own terms.