Online Neural Space Time Memory for Dynamic Novel View Synthesis
The paper proposes updating scene memory periodically, while applying it every frame, to keep dynamic view synthesis real-time.
The method targets multi-view streaming video where occlusions require longer memory, but per-frame test-time training is too costly. It uses cross-view attention to handle deformation between stored memory and the current frame. The authors add Memory Loss and Memory Caching to preserve historical context and limit drift. They report real-time state-of-the-art results on dynamic human motion and minute-scale online memorization. ArXiv · AI/CL/LG's note
The method targets multi-view streaming video where occlusions require longer memory, but per-frame test-time training is too costly. It uses cross-view attention to handle deformation between stored memory and the current frame. The authors add Memory Loss and Memory Caching to preserve historical context and limit drift. They report real-time state-of-the-art results on dynamic human motion and minute-scale online memorization. ArXiv · AI/CL/LG's note
score 4