I will be presenting our new work: “Embodied Scene-aware Human Pose Estimation” at
#NeurIPS2022
Thursday Poster 900.
In this work, we use third person video🎥, proprioception🕺, and scene information🪑 to drive an embodied agent for pose estimation. 1/5
We hypothesize that humans move according to:
1. Movement goal: observed human pose from videos.
2. Proprioception: current agent state including body pose, velocities, etc.
3. Scene awareness: understanding of physical laws and surrounding environments. 2/5
…and propose to use an embodied agent (embodiment = having a tangible existence in a simulated environment) to follow 2D keypoints and “act” inside a simulated environment. 3/5
We show results on *all* of the sequences from the PROX dataset. Our method is *casual*, can run around 10fps (without runtime optimization), and recover realistic human-scene interactions. 4/5