TIGAS streams 3D Gaussian Splatting to thin clients over HTTP/3

TIGAS (Thin-client Interactive Gaussian Adaptive Streaming) tackles the challenge of deploying photorealistic 3D Gaussian Splatting on resource-constrained mobile and XR devices. Traditional 3DGS requires powerful local GPUs for client-side rendering. TIGAS instead moves the heavy rasterization to a remote backend server, then streams lightweight view-dependent 2D projections to a web client over QUIC/HTTP3. This approach eliminates head-of-line blocking and reduces bandwidth demands. A custom adaptive bitrate (ABR) algorithm dynamically adjusts rendering quality based on real-time network conditions, ensuring motion-to-photon latency stays within strict interactive thresholds for 6 degrees-of-freedom (6DoF) experiences.

In extensive evaluations across multi-continental networks using 14 different 3DGS models and real 6DoF EyeNavGS movement traces, TIGAS demonstrated robust performance. The backend server rendered frames in under 10 milliseconds, while the system maintained an average structural similarity index (SSIM) of 0.88. The paper also explores an experimental WebGPU-based super-resolution pipeline, analyzing trade-offs between perceptual quality and thin-client processing limits. By providing an open-source testbed (available on GitHub), TIGAS enables reproducible research into 3DGS streaming and serves as a practical delivery system for mobile XR applications.