The authors pretrained a motion denoiser that represents each rig as a graph, with joints as nodes and bones as edges.
One model generates matched video and motion for arbitrary rigs
JCA couples pretrained video and graph-based motion models, then estimates a camera that places the resulting animation in the depicted scene.
Big Tech
Noam Rotstein · Yoad Tewel · Chen Tessler · Haggai Maron · Gal Chechik · Or Litany · +2 more
NVIDIA · Technion – Israel Institute of Technology · Bar-Ilan University
Research Digest··3 min read
Rotstein and colleagues present JCA, a system that jointly generates video and replayable 3D joint motion for a supplied character rig, including animals, humans and robots.
§