GATOR: Generative and Agentic 3D Object Reconstruction From Casual Images
NVIDIA researchers introduced GATOR, which reconstructs textured, scene-aligned 3D objects from one or more casual images.
NVIDIA researchers presented GATOR, a generative and agentic framework that recovers textured 3D object assets and their scene-relative pose from one or more casual images. A local modality mixer combines patch-aligned RGB, target-mask, and pointmap features, while text prompts such as category names and captions condition structure, geometry, and appearance. The generated asset initializes an agent that refines geometry and texture through an observation-guided edit-render-review loop. The paper reports strong fidelity on synthetic objects, cluttered tabletops, and indoor scenes, plus efficiency and simulation-readiness results.
- GATOR rebuilds textured 3D objects and scene-relative pose from casual images.
- A mixer fuses patch-aligned RGB, masks, and pointmaps before cross-view reasoning.
- Text prompts condition structure, geometry, and appearance via stage adapters.
- A multimodal agent refines the asset through an edit-render-review loop.
From the original article
1 min read at huggingface.co
Reconstructing complete, scene-aligned 3D objects from casual images requires integrating sparse, uncertain observations and inferring surfaces hidden by occlusions. We present GATOR, a generative and agentic framework that recovers textured object assets and their scene-relative pose from one or more images.
By Qirui Wu, Stan Birchfield, Hesam Rabeti · 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.