GUSH3R: Reconstructing Dynamic Humans & Static Scenes from Monocular Video

zhenjun_zhao · x · 2026-07-07

GUSH3R proposes a new method expressing both dynamic humans and static scenes as 3D Gaussians, enabling full-scene 3D reconstruction from a single monocular video. Combining Human3R and the SMPL-X parametric human model, it achieves collaborative 3D modeling of humans and scenes under single-camera video input. This work was published at ICML2026.

Related event: GUSH3R Enables Dynamic 3D Reconstruction from Monocular Video(2 posts)→

Original post →

More from Research

Research channel →