Tetris3D: 3D Scene Generation With Objects That Fit Together
Tetris3D reconstructs single-image 3D scenes so interacting objects remain geometrically and physically coherent.
Tetris3D is a single-image 3D scene generator that explicitly conditions each object's shape and pose on surrounding geometry and physical relationships, rather than generating objects independently. The authors introduce ComOb, a physics-simulation dataset of 1.2 million scenes spanning diverse categories, with per-object meshes and pairwise physical-relation labels. Experiments on synthetic and real scenes show coherent reconstruction of occluded interacting regions and state-of-the-art generation quality and physical stability.
- Conditions each object on neighbors' geometry and physical relations.
- ComOb contains 1.2 million physics-simulated scenes with meshes and pairwise annotations.
- Recovers coherent shapes and poses even when contact regions are occluded.
- Reports state-of-the-art generation quality and physical stability.
- AI models
- Tetris3D
From the original article
1 min read at huggingface.co
We propose Tetris3D, a generative framework for single-image 3D scene reconstruction that recovers objects which are physically and geometrically coherent as a scene. Existing methods often generate objects independently or couple them implicitly, providing limited guidance for ensuring fine-grained spatial compatibility between neighboring objects that interact with one another.
By Jaeyeong Kim, Jinhyuk Jang, Jongmin Lee · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.