Packet-Loss Robust 3D Gaussian Compression via Atomic Packaging and GNN-based Error Concealment
2026-07-20 • Graphics
GraphicsComputer Vision and Pattern RecognitionMultimedia
AI summaryⓘ
The authors address the problem of 3D Gaussian Splatting data being damaged when streamed over networks that lose packets, which causes visual glitches. They create a new method that groups all related data about each 3D point together, so if some packets go missing, the damage is easier to fix by treating it as missing points rather than corrupted details. On the receiving end, they use a smart system combining prior knowledge and graph neural networks to fill in missing information more accurately. Their approach reduces visible errors significantly compared to sending data without protection, even when 20% of packets are lost.
3D Gaussian Splattingpacket lossreal-time neural renderingcompressionerror concealmenthash-grid priorgraph neural networkattribute inpaintingPSNRBungeeNeRF
Authors
Yuxuan Tao, Xuerui Ma, Hao Zhang, Chunhua Peng
Abstract
3D Gaussian Splatting (3DGS) and recent compression schemes such as HAC++ enable high-fidelity real-time neural rendering, but their bitstreams are fragile under packet loss during network streaming. Existing compression methods often separate correlated anchor attributes into independent streams, so losing one packet can create attribute-inconsistent broken anchors and severe rendering artifacts. We propose a packet-loss robust 3DGS transmission and error concealment framework. On the encoder side, anchor-level atomic packaging jointly encapsulates all attributes of each anchor, converting corrupted-attribute failures into clean missing-anchor erasures. Stratified random grouping further disperses packet losses across the spatial domain to avoid large contiguous voids. On the decoder side, we formulate recovery as prior-aware attribute inpainting. A Context-Aware Residual Interpolation (CARI) branch uses hash-grid prior predictions and neighboring residuals to build a robust baseline, while a lightweight two-layer graph neural network with cross-attention over hash-grid priors refines high-frequency attribute residuals. Attribute-wise confidence control falls back to interpolation when learned predictions are unreliable. Experiments under 20 percent random packet loss on BungeeNeRF, Mip-NeRF 360, and Tanks and Temples show that the proposed method substantially improves over no-concealment transmission and limits average PSNR degradation to about 3 dB relative to the lossless HAC++ reference.