POSTER
Fulin Wang; Bo Chen; Lingzhi Zhao; Klara Nahrstedt
3D Gaussian Splatting (3DGS) has emerged as a leading representation for volumetric video. A key challenge in VoD streaming is that low-resolution training views impose an inherent quality ceiling that no compression method can overcome. In this paper, we design GSSR-Stream, a 3DGS VoD streaming system that breaks through this quality ceiling while keeping bandwidth practical. On the server side, GSSR-Stream couples image super-resolution with 3DGS pruning, recovering high-frequency details and then removing redundant Gaussians to control bitrate expansion, and the enhanced key frame are propagated to the remaining frames across the group of frames and compressed. To further reduce bandwidth, GSSR-Stream introduces an intra-viewport pruning strategy that ranks tiles by tile-level Gaussian opacity and projected viewport depth, transmitting visually prominent tiles at full fidelity and less critical tiles at lower coding rates. Experimental results show that GSSR-Stream achieves 83.17%–95.19% bandwidth savings compared to the baseline, saves an additional 6.37%–10.10% bandwidth through intra-viewport pruning, and improves QoE by 61.42%–126.94% under 5G network conditions.