SAM 3D Objects is a foundation model for 3D object reconstruction from images.
The tool reconstructs full 3D shape geometry, texture, and layout from a single image. It addresses the challenge of 3D reconstruction in real-world scenarios by handling occlusion and clutter through progressive training and a data engine incorporating human feedback. Given a masked object in an image, the model outputs a 3D representation with pose, shape, texture, and layout information. The approach is designed to work robustly on small objects, unusual poses, and the kinds of uncurated natural scenes that challenge existing 3D generation methods.
Developers working on 3D reconstruction pipelines, particularly those needing to process real-world images with occlusions and complex scenes, should consider this tool. The project provides code, model weights, and a web demo for experimentation. A benchmark is also available for evaluation. The tool is part of a broader SAM 3D effort; a separate model for human mesh reconstruction exists as a companion project.
The project maintains a substantial base of adopters who report issues from real-world use cases rather than the core team generating most open issues. Maintainers typically respond to new issues and pull requests within a day.