Robust and efficient cross-device 3D Gaussian Splatting in web browsers

Establish robust and efficient rendering techniques for 3D Gaussian Splatting that achieve consistent performance and correct synchronization across heterogeneous web browsers and GPU hardware, enabling reliable cross-device deployment on the web.

Background

WebGPU introduces modern GPU compute capabilities to the browser, but naive ports of native CUDA/Vulkan algorithms often assume scheduling guarantees or rely on global atomics that are not available or consistent in WebGPU. The web ecosystem is highly heterogeneous, spanning discrete GPUs, integrated graphics, and mobile SoCs, which can lead to performance degradation and synchronization failures.

Because of this diversity and the mismatch between traditional graphics pipelines and neural rendering workloads like depth sorting and view-adaptive shading, achieving robust and efficient 3D Gaussian Splatting rendering across browsers and devices is not straightforward and has been identified by the authors as a significant open problem.