MVLift estimates global 3D human, animal, and interaction motion from single-view 2D poses using stepwise multi-view 2D diffusion and epipolar geometry, without end-to-end 3D supervision.
Beyond static features for temporally consistent 3d human pose and shape from a video
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Lifting Motion to the 3D World via 2D Diffusion
MVLift estimates global 3D human, animal, and interaction motion from single-view 2D poses using stepwise multi-view 2D diffusion and epipolar geometry, without end-to-end 3D supervision.