Causal Spatiotemporal Mamba Enables Feed-Forward Monocular 4D Reconstruction
zhenjun_zhao · x · 2026-10-23
The paper proposes a feed-forward framework for causal 4D visual geometry learning from monocular videos, addressing unordered temporal aggregation and disjoint geometry/motion supervision in dynamic scene reconstruction.
Key components:
- Causal Spatiotemporal Mamba (CSM): augments temporal aggregation with directional state propagation; complementary Temporal-Spatial and Spatial-Temporal Scans continuously propagate geometry-aware states across frames to model spatiotemporal evolution.
- Unified Motion Supervision (UMS): instead of treating motion as an independent objective, UMS uses geometry-aware predictions from camera and depth heads to build a unified forward motion field consistent with the underlying 3D structure, explicitly constraining inter-frame state transitions for consistent 4D scene representations.
More from Research
- Berkeley's StyleStream 2.0 cuts streaming voice conversion latency to 520ms, 2x faster — Berkeley · 2026-10-23
- Math PhD student quits academia: AI is making math unrecognizable — pickover · 2026-10-23
- ICLR Submission Found Containing LLM's Own Meta-Comment Admitting Results Weren't Significant — _onionesque · 2026-10-23
- AKV Serving System Lets Agents Manage Their Own KV Cache, Cutting Cost 27.7% — _akhaliq · 2026-10-23
- AuroraTTS-100K: a sub-100K-parameter English TTS running 1120x real-time on GPU — Tall_Abrocoma_3533 · 2026-10-23
- MiTaS multi-resolution tactile imitation learning hits 80% success vs 35.5% vision-only baseline — Jan_R_Peters · 2026-10-23