Get Started
Home
Topics
Search
Library
Research questionHow can online novel-view synthesis reconstruct scenes causally from streaming observations while staying renderable and efficient?The system must update a scene reconstruction as observations arrive rather than wait for a complete recording, while producing usable novel views throughout processing. Predicted camera poses and growing sequence context can make reconstruction stability and runtime difficult.
AI
Computer Vision
Image & Video Processing
Inference Optimization
Research Paper
Technology
Latest papersRecent research connected to this question, newest first.ReCoSplat: Online Feed-Forward Gaussian Splatting via Render-and-CompareApplies to the ReCoSplat system and its online Gaussian Splatting task. The supplied evidence covers 256-view streams, reported input and end-of-stream throughput, and KV-cache compression on long sequences; it does not establish broader deployment performance.research paper · Sep 2, 2026
Related questions
How can feed-forward 3D Gaussian Splatting meet a fixed Gaussian budget without sacrificing novel-view quality?How can Gaussian splatting render unseen viewpoints reliably when large-scale scenes have incomplete, unstructured view coverage?How can dynamic 4D Gaussian splatting reduce per-scene storage without sacrificing view-synthesis quality or real-time rendering?How can sparse-view 3D scene representations avoid cross-view artifacts without costly per-scene reconstruction?