Revisiting Local Context for Long-Horizon Streaming 3D Reconstruction
- 类型：arxiv
- 标识：2608.27529
- 链接：https://arxiv.org/abs/2608.27529
- 主分类：llm-infra
- 形态：method
- TLDR：Streaming 3D reconstruction from extremely long videos requires estimating camera motion and scene geometry online under bounded memory and computation. Early streaming models achieve causal, bounded-cost inference using finite context buffers or compact recurrent states, yet their estimates often deteriorate as sequences grow. Recent methods improve long-horizon stability by coupling short-range context with persistent or multi-level long-range memory. We pursue a different route: we keep the learned temporal state strictly local and formulate predictions whose targets remain independent of s
- 待LLM分类：否
- 来源文件：
- /inbox/tom/_candidates/2026-08-31-agent-rag-longcontext-candidates.json