3DGS Acceleration

The field studies how to make 3D Gaussian Splatting faster and smaller for tasks such as novel-view rendering, scene reconstruction, and mobile or city-scale graphics. Approaches include faster rasterization and tile scheduling, culling and pruning redundant Gaussians, level-of-detail and compression methods, and training changes such as sparse sampling or adaptive densification. Hardware-aware kernels, distributed rendering, and alternative transparency or blending methods target bottlenecks from GPU execution to memory and communication. The main tension is balancing speed and model size against visual fidelity, especially for transparency, fine detail, and varied scenes and devices.