BAT3R: Bayesian Adaptive State Updates Stabilize Streaming 3D Reconstruction Over 1000 Frames
zhenjun_zhao · x · 2026-10-21
BAT3R tackles error accumulation in streaming 3D reconstruction with a training-free Bayesian adaptation framework. State Innovation Gating anchors state evolution to its geometric initialization to arrest representational drift, while Bayesian-guided Adaptive State Updating reformulates updates as recursive Bayesian inference within a Kalman framework, modeling temporal reliability via information-theoretic gain and spatial relevance via cross-attention — protecting consolidated regions from redundant updates. On ScanNet and TUM RGB-D, pose error grows far more slowly with view count while baselines OOM; BAT3R scales stably to 1000 input frames.
More from Research
- Domingos Cites New Result Claiming RL Is 'Basically Useless' vs Supervised Learning — pmddomingos · 2026-10-23
- Wolfram Opens Rule 30 Prizes to AIs: $30,000 for Three Open Problems — cephaloform · 2026-10-23
- Stanford and Databricks use AI agents to reproduce medical findings and test analytical robustness — ChengleiSi · 2026-10-23
- EMNLP paper: diffusion LMs internally encode latent denoising timestep representations — gsarti_ · 2026-10-23
- KU Leuven and VIB open two Group Leader professorships in AI for Biology, start 2027 — jpillowtime · 2026-10-23
- Subspace Tuning curbs LLM catastrophic forgetting by freezing principal singular directions — scottniekum · 2026-10-23