VIDAR: Visual-Inertial Dense Alignment and Reconstruction via a Geometric Foundation Model
Robotics
Summary
The authors present VIDAR, a system that combines camera movement data and a special depth model to create detailed 3D maps from a single camera. They use measurements from both the camera and an inertial sensor to keep track of the real-world scale and position, solving a common problem where single-camera models can't measure true size. Their method aligns detailed depth predictions over time into one consistent map. Tests show that VIDAR significantly improves the accuracy and detail of 3D reconstructions compared to previous methods.
Authors
Diyari Mohammed Salih, Lingxiang Hu, Naima AitOufroukh-Mammar, Fabien Bonardi
Abstract
Monocular foundation models provide dense geometry but usually lack a stable metric scale. This paper presents VIDAR, a visual-inertial dense reconstruction framework that couples SVO+IMU odometry with Depth Anything 3. VIDAR uses the visual-inertial front end as a metric anchor: it provides camera poses, scale, and a consistent world frame for aligning dense foundation-model predictions across time. The foundation model then contributes detailed local geometry that is fused into a global reconstruction. We study both pose-conditioned DA3 and a decoupled alignment strategy. On EuRoC, pose injection reduces scale error to about 1\% and reaches 0.463 mean F@0.10; the decoupled hybrid improves this to 0.676 without ground-truth poses. Results on EuRoC and TUM RGB-D show that VIDAR is a practical route to metric dense monocular reconstruction.