SAM 3D: 3Dfy Anything in Images

TL;DR

SAM 3D introduces a scalable, multi-stage framework for 3D object reconstruction from a single image, leveraging human-in-the-loop annotations and synthetic pretraining to outperform recent methods.

Contribution

The paper presents SAM 3D, a novel generative model that combines human and model-in-the-loop annotation with a multi-stage training process to improve 3D reconstruction from images.

Findings

Achieves at least a 5:1 preference win over recent methods.

Provides a new large-scale dataset for in-the-wild 3D reconstruction.

Outperforms prior work in natural, cluttered scenes.

Abstract

We present SAM 3D, a generative model for visually grounded 3D object reconstruction, predicting geometry, texture, and layout from a single image. SAM 3D excels in natural images, where occlusion and scene clutter are common and visual recognition cues from context play a larger role. We achieve this with a human- and model-in-the-loop pipeline for annotating object shape, texture, and pose, providing visually grounded 3D reconstruction data at unprecedented scale. We learn from this data in a modern, multi-stage training framework that combines synthetic pretraining with real-world alignment, breaking the 3D "data barrier". We obtain significant gains over recent work, with at least a 5:1 win rate in human preference tests on real-world objects and scenes. We will release our code and model weights, an online demo, and a new challenging benchmark for in-the-wild 3D object reconstruction.