学术研究★★★arXiv · 2026-07-16
Online Neural Space-Time Memory for Dynamic Novel View Synthesis
Online novel view synthesis from multi-view streaming videos requires balancing persistent long-term memory with real-time constraints. Test-Time Training offers a robust memory mechanism but standard models necessitate gradient-based updates per frame, leading to high computational costs and instability.
📌 Key points
- Balancing persistent memory and real-time constraints is crucial for online nove
- Test-Time Training provides a powerful memory mechanism
- Standard models require costly gradient updates per frame leading to instability
本页为 gitzw.com 基于公开来源的 AI 中文解读，非原文转载。