Tetris3D: 3D Scene Generation With Objects That Fit Together
Tetris3D conditions each generated object on its neighbors so the reconstructed scene holds together physically and geometrically.
The paper targets single-image 3D scene reconstruction where nearby objects interact, including cases where contact regions are occluded. It adds explicit conditioning on surrounding geometry and physical relationships to guide both shape and pose. The authors also introduce ComOb, a 1.2M-scene physics-simulation dataset with per-object meshes and pairwise relation annotations. They report state-of-the-art generation quality and physical stability on synthetic and real-world scenes. HF Daily Papers' note
The paper targets single-image 3D scene reconstruction where nearby objects interact, including cases where contact regions are occluded. It adds explicit conditioning on surrounding geometry and physical relationships to guide both shape and pose. The authors also introduce ComOb, a 1.2M-scene physics-simulation dataset with per-object meshes and pairwise relation annotations. They report state-of-the-art generation quality and physical stability on synthetic and real-world scenes. HF Daily Papers' note
score 5