KAIST's Tetris3D Reconstructs 3D Scenes With Physically Coherent Objects, Plus 1.2M-Scene ComOb Dataset
kaist-ai · hf · 2026-10-08
KAIST AI introduces Tetris3D, a generative framework for single-image 3D scene reconstruction that enforces fine-grained spatial compatibility between interacting objects.
- Key idea: each object's generation is explicitly conditioned on the geometry and physical relationships of surrounding objects, keeping shapes and poses physically plausible even when interaction regions are occluded.
- Companion dataset ComOb: 1.2M physics-simulated scenes across diverse object categories, with per-object meshes and pairwise physical relation annotations.
- Experiments on synthetic and real-world scenes show state-of-the-art generation quality and physical stability.
More from Multimodal
- Kling 4.0 Flash demo: one-prompt Christmas video shared by AI creator — umesh_ai · 2026-10-08
- LTX Alpha Matte LoRA + H3 Inpainting enable surprisingly good video background replacement — linoy_tsaban · 2026-10-08
- New tool lets you inspect PS1 and N64 level UVs and textures side by side with live 3D models — Promptmethus · 2026-10-08
- Person Remover LoRA for MiniMax H3 Released with Full ComfyUI Workflow — linoy_tsaban · 2026-10-08
- Suno Praised as Good Enough to Eventually Overtake Spotify — petergyang · 2026-10-08
- 3.5M-param anime upscaler applies Looped-DiT recurrence, gains +0.6dB in 4 loops — NobodySnJake · 2026-10-08