CaT-GS: Efficient 3DGS Rendering for Large-Scale Scenes with Inter-frame Caching and Tile Scheduling

Recent breakthroughs in 3D Gaussian Splatting (3DGS) have advanced neural rendering with high fidelity and efficiency. However, its performance degrades severely in large-scale scenes due to the increasing computational workload of tile-based rasterization. Existing acceleration approaches either require costly scene re-training or focus only on the rasterization stage of the pipeline, overlooking general pipeline redundancy in real-time rendering. Through a comprehensive analysis, we identify three primary sources of redundancy and low GPU utilization: 1) redundant inter-frame pre-processing, 2) viewpoint redundancy, and 3) imbalanced tile load distribution. To address these issues, we propose CaT-GS, a novel and efficient 3DGS rendering pipeline. CaT-GS introduces a speculative multi-frame preprocessing method to eliminate redundant computations across consecutive frames, and an interframe caching mechanism to eliminate viewpoint redundant rendering stages. Furthermore, it redistributes the rasterization tasks with a dedicated CUDA kernel to mitigate tile load imbalance and boost GPU utilization. Extensive experiments on large-scale scenes demonstrate that CaT-GS achieves a speedup of up to 10! over the original 3DGS and up to 70% over previous state-of-the-art (SOTA) methods, establishing a new benchmark for high-fidelity, realtime rendering of large-scale scenes.