PoseGPT: Quantizing human motion for large scale generative modeling
Thomas Lucas, Fabien Baradel, Philippe Weinzaepfel, Gregory Rogez
Summary
We address the problem of action-conditioned generation of human motion sequences.
Unlike existing work, we generate motion conditioned on observations of arbitrary length, including none. To solve this generalized problem, we propose PoseGPT, an auto-regressive transformer-based approach which internally compresses human motion into quantized latent sequences. Inspired by the Generative Pretrained Transformer (GPT), we propose to train a GPT-like model for next-index prediction in that space; this allows PoseGPT to output distributions on possible futures, with or without conditioning on past motion. We mainly experiment on BABEL, a recent large scale MoCap dataset.
Unlike existing work, we generate motion conditioned on observations of arbitrary length, including none. To solve this generalized problem, we propose PoseGPT, an auto-regressive transformer-based approach which internally compresses human motion into quantized latent sequences. Inspired by the Generative Pretrained Transformer (GPT), we propose to train a GPT-like model for next-index prediction in that space; this allows PoseGPT to output distributions on possible futures, with or without conditioning on past motion. We mainly experiment on BABEL, a recent large scale MoCap dataset.
