R3D3: Dense 3D Reconstruction of Dynamic Scenes from Multiple Cameras

TL;DR

R3D3 introduces a multi-camera system that combines geometric estimation and learned scene priors to achieve dense, consistent 3D reconstructions of dynamic outdoor scenes, outperforming existing methods on key benchmarks.

Contribution

The paper presents a novel multi-camera approach with integrated geometric and learned priors for improved 3D reconstruction of dynamic scenes.

Findings

State-of-the-art dense depth prediction on DDAD benchmark.

Robust geometric depth and pose estimates in challenging environments.

Effective reconstruction of moving objects and low-texture regions.

Abstract

Dense 3D reconstruction and ego-motion estimation are key challenges in autonomous driving and robotics. Compared to the complex, multi-modal systems deployed today, multi-camera systems provide a simpler, low-cost alternative. However, camera-based 3D reconstruction of complex dynamic scenes has proven extremely difficult, as existing solutions often produce incomplete or incoherent results. We propose R3D3, a multi-camera system for dense 3D reconstruction and ego-motion estimation. Our approach iterates between geometric estimation that exploits spatial-temporal information from multiple cameras, and monocular depth refinement. We integrate multi-camera feature correlation and dense bundle adjustment operators that yield robust geometric depth and pose estimates. To improve reconstruction where geometric depth is unreliable, e.g. for moving objects or low-textured regions, we introduce learnable scene priors via a depth refinement network. We show that this design enables a dense, consistent 3D reconstruction of challenging, dynamic outdoor environments. Consequently, we achieve state-of-the-art dense depth prediction on the DDAD and NuScenes benchmarks.