Research map / Architectures and efficiency / Attention and transformers
FILT3R: Latent State Adaptive Kalman Filter for Streaming 3D Reconstruction
Seonghyun Jin, Jong Chul Ye
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace uniform overwrite/heuristic gate with Kalman-style adaptive filtering (FILT3R) | CUT3R backbone · 7-Scenes (length 1000) | Acc mean: 0.145 -> 0.054 (TTT3R vs FILT3R) | Improved |
| replace TTT3R gating with FILT3R | CUT3R backbone · 7-Scenes (length 300 and 500) | Acc mean: 0.040 to 0.020 at length 300 and 0.066 to 0.024 at length 500 | Improved |
| replace TTT3R/CUT3R update with FILT3R | CUT3R backbone · TUM-800 | ATE_orig: 0.214 -> 0.107 | Improved |
| replace TTT3R with FILT3R | CUT3R backbone · TUM-800 | RPE-t: 0.009 | No clear effect |
| replace TTT3R with FILT3R | CUT3R backbone · Bonn-300 (metric depth) | Abs Rel: 0.108 to 0.093 | Improved |
| add periodic hard reset to TTT3R instead of FILT3R | TTT3R · TUM-800 | ATE_orig: 0.107 vs 0.221 | Hurt |
| add drift-driven adaptive process noise (vs fixed Q) (ablation) | FILT3R · TUM-800 | ATE_orig: 0.229 -> 0.107 | Improved |
| add drift-driven adaptive process noise (vs fixed Q) (ablation) | FILT3R · Bonn-500 (metric depth) | Abs Rel: 0.099 -> 0.089 | Improved |
| add variance propagation (vs reset-P) (ablation) | FILT3R · TUM-800 | ATE_orig: 0.489 -> 0.107 | Improved |
| use Kalman gain instead of tuned fixed-beta EMA (0.05) (ablation) | FILT3R · TUM-800 | ATE: 0.049 vs 0.057; RPE-r 0.658 -> 0.362 | Mixed |
| add EMA normalization of drift (ablation) | FILT3R · TUM-800 | ATE_orig: 0.109 -> 0.107 | No clear effect |
| use fixed scalar measurement noise instead of adaptive r_t (ablation) | FILT3R · TUM-800 | ATE_orig: 0.148 -> 0.107 | Improved |
Closest papers by meaning
- BAT3R: Robust Online 3D Reconstruction with Bayesian Adaptive State Updates (NeurIPS 2026 · similarity 0.92)
- Stable3R: Streaming 3D Reconstruction with Stable Geometric References (NeurIPS 2026 · similarity 0.90)
- Rethinking the State Update Gate for Long-Sequence Recurrent 3D Reconstruction (NeurIPS 2026 · similarity 0.90)
- Continuous 3D Perception Model with Persistent State (CVPR 2025 · similarity 0.88)
- MESSENGER: Memory-Enhanced Sequential Scene Flow Estimation via Autoregressive Next-Frame Forecasting (NeurIPS 2026 · similarity 0.87)
- Living Scenes: Multi-object Relocalization and Reconstruction in Changing 3D Environments (CVPR 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
