BLASt3R: Bundle Adjustment of Any Image Set with Multi-View Matching and Monocular Priors

On September 4, 2026, the BLASt3R system was released in the arXiv preprint cs.CV domain. This system is a normalized beam adjustment framework that combines multi-view matching with single-eye priors. Its core feature is that it supports both online visual SLAM (VSLAM) and offline reconstruction of unordered image sets within a unified optimization framework, and shares hyperparameters for all tasks. Experimental results show that BLASt3R outperforms traditional, feedforward, and hybrid baseline methods in terms of performance and speed; especially in the VSLAM domain, its non-calibrated method performs better than all previous calibration schemes.