VidMap: Exploiting Temporal Structure for Video-Based Structure-from-Motion

Researchers introduced VidMap, a video-based structure-from-motion system for metric reconstruction from arbitrary long uncalibrated videos.

VidMap combines SLAM-like temporal constraints with global SfM optimization, wide-baseline dense matching, and monocular depth priors.

On challenging datasets, it outperformed prior SLAM and SfM methods in accuracy and robustness.

The approach could improve camera pose estimation and 3D scene reconstruction for navigation and scene-understanding data generation.