Rig3R

Overview

Rig3R is a geometric foundation model developed by Wayve for robust ego-motion and 3D structure estimation from multi-camera rigs. It leverages rig constraints when available and infers them when missing, achieving state-of-the-art 3D understanding in complex driving scenarios. The model processes multiple frames and views in a single forward pass, outperforming baselines by 17–45% on real-world benchmarks.

Key features

• ▸First learned method to explicitly leverage rig constraints for multi-camera 3D reconstruction
• ▸Processes multiple frames and camera views in a single forward pass
• ▸Outperforms traditional and learned methods by 17–45% on real-world driving benchmarks
• ▸Robust to missing or incomplete calibration
• ▸Supports dense 3D reconstruction, camera pose estimation, and rig calibration