A GS-Cache Inference Framework for Large-Scale Gaussian Splatting Models

In this study, the authors address the challenges of rendering large-scale 3D Gaussian Splatting (3DGS) models on consumer-grade devices in real-time. They introduce GS-Cache, an innovative framework that integrates advanced 3DGS representation with a highly optimized rendering system to support immersive experiences, such as virtual reality (VR). GS-Cache improves performance, reduces latency, and lowers GPU memory usage, enabling high-quality 2K binocular rendering at over 120 FPS. This novel approach bridges the gap between 3DGS’s capabilities and the requirements of VR systems, establishing a scalable and efficient framework for real-time neural rendering in immersive environments.
Controversial information: None
Surprising content: Achieving 2K binocular rendering at over 120 FPS.

https://arxiv.org/abs/2502.14938

To top