DynStream Delivers Online 4D Gaussian Reconstruction from Unposed Video Streams
zhenjun_zhao · x · 2026-10-09
DynStream: Streaming 4D Gaussian Reconstruction from Unposed Video
- Background: Existing feed-forward Gaussian methods are offline-only; online point-cloud approaches struggle with dense geometry and photorealistic rendering.
- Method: Reconstructs scenes within local temporal windows, then incrementally aligns and fuses them into a globally consistent scene without per-scene optimization, enforcing cross-window geometric consistency and modeling time-varying content.
- Results: State-of-the-art high-fidelity online dynamic reconstruction across diverse indoor and outdoor scenes. Authors include Ming-Hsuan Yang.
More from Research
- Are URM and Universal Transformers the forgotten architecture beating standard LLMs? — moschles · 2026-10-09
- Researcher: Use AI to Rewrite Machine-Generated Math Proofs Into Human-Readable Forms — jd_pressman · 2026-10-09
- AutoScientist's two-agent checklist loop auto-audits every training example — sarahookr · 2026-10-09
- NeurIPS GenAI4Health oral: retrieval-based medical fact-checking fails in ways bigger models can't fix — mdredze · 2026-10-09
- Bigger models, more reasoning, better sources won't fix medical fact-checking, researchers say — mdredze · 2026-10-09
- DEX best abstract: top LLMs catch many physician diagnostic errors, but big gaps remain — mdredze · 2026-10-09