Search papers, labs, and topics across Lattice.
2
0
2
Forget expensive 3D human datasets: DrPose directly fine-tunes multi-view diffusion models using only single-view images and human poses, leading to more natural 3D human reconstructions, particularly for dynamic poses.
Ditch the hand-crafted priors: SEAL-pose learns a data-driven loss function that significantly boosts 3D human pose estimation by enforcing structural consistency, outperforming models with explicit constraints.
