F²SLAM Turns Feed-Forward 3D Geometry into Persistent Factors, Cutting SLAM Error by an Order of Magnitude

Researchers including Javier Civera propose F²SLAM, which converts feed-forward 3D model predictions into persistent factors in SLAM optimization, reducing error by an order of magnitude. On Replica in uncalibrated settings, average ATE drops to 0.002m, with results also reported on TUM RGB-D, 7-Scenes, and KITTI.

2026-10-19 ~ 2026-10-19 · 3 related posts

• F²SLAM Turns Feed-Forward Geometry into Persistent Factors, Cutting Replica ATE to 0.002 m — zhenjun_zhao · 2026-10-19
• F²SLAM Benchmarks: 0.002 m ATE on Uncalibrated Replica at 7.4 FPS on One A100 — zhenjun_zhao · 2026-10-19

1 near-duplicate retellings: zhenjun_zhao