Glad to share the main research of my master's degree, VideoMDM! Please go check it out! videomdm.github.io
videomdm.github.io VideoMDM — 3D Human Motion Generation from 2D Supervision A diffusion-based framework that trains 3D human motion priors from monocular video, without any 3D ground truth. Excited to share our new paper: VideoMDM 📢
We propose a principled framework for training 3D motion diffusion models (e.g. MDM), using only 2D supervision from monocular videos -- no 3D ground truth required.
Project: videomdm.github.io
Paper: arxiv.org/abs/2606.13364
🧵View quoted post on X
