VIBE: Estimating Human Motion with AI #ArtificialIntelligence #MachineLearning #PyTorch #AdversarialNetworks @MPI_IS
At the end of last year, the Max Planck Institute for Intelligent Systems published a paper on VIBE – Video Inference for Human Body Pose and Shape Estimation. VIBE is a machine learning framework that is able to digest video and produce motion sequences without 3D labels. The framework has a number of moving pieces and utilizes data from the AMASS motion capture dataset.
Our key novelty is an adversarial learning framework that leverages AMASS to discriminate between real human motions and those produced by our temporal pose and shape regression networks. We define a temporal network architecture and show that adversarial training, at the sequence level, produces kinematically plausible motion sequences without in-the-wild ground-truth 3D labels.
The code for the PyTorch implementation of VIBE can be found on @mkocabas‘ GitHub Repo. If you’d like to try out the implementation there is also a CoLab version. If you’d like to see some example output check out this video.