The authors divide an unposed image sequence into two interleaved temporal subsets and build an independent Gaussian submap for each.
Cross-validation between scene maps reduces drift in pose-free 3D reconstruction
CrossFit-GS estimates camera trajectories by refining each frame against a separate map that was not trained on that frame.
Top University
Zijian Wu · Jinliang Wang · Mingfeng Jiang · Zidian Lin · Ying Song · Ziqian Lu
Zhejiang Sci-Tech University · Tsinghua University
Research Digest··3 min read
Wu et al.
§