GS-NFS: Bandwidth-adaptive Streaming of Dynamic Gaussian Splats and Point Clouds

Amartya Chaudhuri; Antonio Ortega; Eduardo Pavez; Haodong Wang; Haoran Hong; Harsha V. Madhyastha; Rajrup Ghosh; Ramesh Govindan; Weiwu Pang

arxiv: 2606.05650 · v1 · pith:JFMNZQA4new · submitted 2026-06-04 · 💻 cs.MM · cs.CV· cs.GR· cs.NI

GS-NFS: Bandwidth-adaptive Streaming of Dynamic Gaussian Splats and Point Clouds

Rajrup Ghosh , Haodong Wang , Haoran Hong , Eduardo Pavez , Amartya Chaudhuri , Weiwu Pang , Harsha V. Madhyastha , Antonio Ortega

show 1 more author

Ramesh Govindan

This is my paper

Pith reviewed 2026-06-27 23:03 UTC · model grok-4.3

classification 💻 cs.MM cs.CVcs.GRcs.NI

keywords dynamic 3D Gaussian splattingGPU compression3D video streamingpoint cloud encodingreal-time decompressionbandwidth adaptive

0 comments

The pith

A GPU method parallelizes 3D Gaussian splat encoding to run 10-100 times faster than prior work.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents GS-NFS, which applies novel GPU parallelizations to existing algorithms for compressing dynamic 3D Gaussian Splatting frames. Prior methods for handling the large per-frame data in 3DGS video were too slow for real-time use, even though the representation itself supports high-fidelity arbitrary-view rendering. A reader would care if the speedup holds because it could make bandwidth-adaptive 3D video streaming practical at full frame rates. The work shows encoding and decoding speeds one to two orders of magnitude above the state of the art while keeping compression ratios and rendered quality competitive.

Core claim

GS-NFS accelerates dynamic 3DGS compression and decompression on a GPU by developing novel GPU-based parallelizations of existing algorithms for encoding both positions and attributes of Gaussians. As a result it encodes and decodes a frame 1-2 orders of magnitude faster than the state-of-the-art while offering competitive compression performance and rendering quality.

What carries the argument

Novel GPU-based parallelizations of existing algorithms for encoding positions and attributes of Gaussians.

If this is right

Dynamic 3DGS content can be encoded and decoded at full frame rate.
The approach supports bandwidth-adaptive streaming of 3D scenes represented as Gaussians.
Compression ratios remain competitive with earlier non-GPU methods.
Rendered output quality stays comparable to uncompressed or prior compressed versions.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same parallelization pattern could be tested on static 3DGS or raw point-cloud streams to check for similar speedups.
Integration with existing 2D video pipelines might become feasible once per-frame 3DGS data moves at video rates.
Mobile or edge devices could become practical endpoints for 3DGS streams if the GPU kernels are further adapted to lower-power hardware.

Load-bearing premise

The GPU parallelizations of the position and attribute encoding steps preserve both the claimed speed gains and the original compression ratios without hidden quality or correctness trade-offs that only appear in full evaluation.

What would settle it

A timing measurement on standard hardware in which GS-NFS encoding or decoding of a typical dynamic 3DGS frame takes longer than one-tenth the time of the previous best method, or a side-by-side rendering test showing quality below the competitive threshold reported.

Figures

Figures reproduced from arXiv: 2606.05650 by Amartya Chaudhuri, Antonio Ortega, Eduardo Pavez, Haodong Wang, Haoran Hong, Harsha V. Madhyastha, Rajrup Ghosh, Ramesh Govindan, Weiwu Pang.

**Figure 1.** Figure 1: GS-NFS Architecture. Blocks in yellow are GPU-resident components. During-training compression techniques reduce information in a variety of ways. QUEEN [10], 4DGC-Pro [49], and CompGS++ [23] reduce the number of Gaussians by using temporal prediction between key frames and their immediately following frames. A frame following a key frame is represented using residuals—differences with respect to the key… view at source ↗

**Figure 2.** Figure 2: An example of an octree with some occupied voxels. Then, it maps each Gaussian’s 3-D position to one of the cubes, by quantizing the position coordinates to use 𝐽 bits. Multiple Gaussians might fall into a single voxel; GS-NFS replaces those Gaussians with a single Gaussian whose position is at the center of the voxel and other attributes are the average of the original Gaussians.2 The Octree Data Structur… view at source ↗

**Figure 3.** Figure 3: This figure shows the Morton codes of the occupied voxels in [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

**Figure 4.** Figure 4: Example to illustrate the RAHT algorithm. to voxels, and in [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

**Figure 5.** Figure 5: Heatmaps that plot Pearson correlation across SHs in RGB domain (left) and after YUV conversion and KLT (right). heatmap of the Pearson correlation coefficients for the original RGB SH coefficients. On the right is the corresponding heatmap after applying RGB to YUV conversion followed by KLT; notice how the figure on the right has significant white-space (white corresponds to zero correlation). By decorr… view at source ↗

**Figure 6.** Figure 6: R-D curves for (Left) Actor1 from HiFi4G, (Middle) flame_salmon from N3DV, and (Right) sear_steak from N3DV. 5.3 Rate-Distortion Curves To understand quality and compression trade-offs of different designs, we can also use a rate-distortion curve or R-D curve. Such a curve plots rate on the x-axis, quality or distortion on the y-axis, and the curve represents the best quality one can obtain for a given rat… view at source ↗

read the original abstract

Dynamic 3D Gaussian Splatting (3DGS) holds great promise as a 3D video streaming technology since it can represent complex 3D scenes with high fidelity. In this approach, every frame in a 3D video represents the environment as a collection of Gaussians with position and other attributes such as scale, rotation, opacity, and color. Frames capture fine details, permit views from any arbitrary perspective, but are an order of magnitude, or more, larger than 2D video frames. A line of recent work has explored how to compress dynamic 3DGS frames, but these approaches are often slow, in part because their compression techniques are not amenable to efficient acceleration. GS-NFS accelerates dynamic 3DGS compression and decompression on a GPU, to the point where it can encode and decode at full frame rate. It achieves this by developing novel GPU-based parallelizations of existing algorithms for encoding both positions and attributes of Gaussians. As a result, it is 1-2 orders of magnitude faster than the state-of-the-art in encoding and decoding a frame, while offering competitive compression performance and rendering quality.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

GS-NFS applies GPU parallelizations to speed up dynamic 3DGS encoding and decoding to full frame rates, but the abstract gives no implementation details or numbers to check the claims.

read the letter

The main takeaway is that this paper builds a system to compress and stream dynamic 3D Gaussian splats at interactive rates by rewriting position and attribute encoding steps for GPU parallelism. The abstract states this yields 1-2 orders of magnitude faster encode/decode than prior work while holding compression and quality steady.

What stands out is the focus on a real bottleneck: recent dynamic 3DGS methods are accurate but too slow for live streaming. The work targets that directly by adapting existing algorithms to run in parallel on GPU, aiming for full frame rate operation in bandwidth-limited settings like VR/AR.

The approach reads as straightforward engineering that could matter if the speedups materialize without quality loss. It correctly identifies that current compression techniques do not map well to fast hardware.

The soft spot is the missing substance. The abstract supplies no equations, no description of the parallelization strategy, no per-metric deltas, and no baseline tables. Without those, it is impossible to tell whether the claimed speedups come from genuine advances or routine GPU porting, or whether hidden trade-offs appear at scale. The stress-test note flags the same gap.

This paper is aimed at people building 3D video pipelines and compression tools. A reader who needs practical streaming for dynamic scenes would want the full results. It deserves peer review because the problem is concrete and the direction is relevant, even though the current write-up stays high-level.

Referee Report

1 major / 0 minor

Summary. The paper introduces GS-NFS, a GPU-accelerated framework for bandwidth-adaptive streaming of dynamic 3D Gaussian Splats (and point clouds). It develops novel parallelizations of existing position and attribute encoding algorithms to enable full frame-rate encode/decode, claiming 1-2 orders of magnitude speedup over prior art while maintaining competitive compression ratios and rendering quality.

Significance. If the speedup and quality claims are substantiated with concrete metrics, the work would be significant for enabling practical real-time dynamic 3D content delivery in VR/AR and 3D video applications, where frame sizes currently limit deployment.

major comments (1)

Abstract: the central claim of '1-2 orders of magnitude faster than the state-of-the-art in encoding and decoding a frame' is presented without any quantitative results, baselines, error bars, or per-metric deltas. This is load-bearing for the contribution, as the speedup is the primary asserted advantage and cannot be assessed from the given text.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their review. We address the single major comment below and will revise the manuscript accordingly.

read point-by-point responses

Referee: Abstract: the central claim of '1-2 orders of magnitude faster than the state-of-the-art in encoding and decoding a frame' is presented without any quantitative results, baselines, error bars, or per-metric deltas. This is load-bearing for the contribution, as the speedup is the primary asserted advantage and cannot be assessed from the given text.

Authors: We agree that the abstract should include concrete quantitative support for the speedup claim to allow readers to assess it independently. The results section already reports specific per-metric speedups (encoding and decoding), baselines, and quality metrics with standard deviations. In the revision we will add the key numerical deltas and baselines directly into the abstract while preserving its length and readability. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper describes an engineering system for GPU-based parallelization of existing compression algorithms for dynamic 3D Gaussian Splatting. No derivation chain, equations, predictions, fitted parameters, or first-principles results are presented in the abstract or summary. The central claims concern implementation speedups and competitive quality, which rest on empirical evaluation rather than any self-referential mathematical construction. The contribution is self-contained as a systems paper with no load-bearing steps that reduce to inputs by definition or self-citation.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no information on free parameters, axioms, or invented entities.

pith-pipeline@v0.9.1-grok · 5777 in / 946 out tokens · 14216 ms · 2026-06-27T23:03:38.682386+00:00 · methodology

discussion (0)

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

Renderable Partial Representations for Dynamic Gaussian Splatting under Incomplete Delivery
cs.GR 2026-06 unverdicted novelty 7.0

Dynamic Gaussian representations are clustered into spatiotemporal groups with render-utility-optimized refinements so that partial deliveries remain directly renderable, removing PSNR regressions and gaining 3.03 dB ...

Reference graph

Works this paper leans on

53 extracted references · 19 canonical work pages · cited by 1 Pith paper

[1]

[n. d.]. Draco 3D Compression. https://github.com/google/draco
[2]

[n. d.]. Point Cloud Library. https://pointclouds.org/
[3]

2012.Space-filling curves: an introduction with appli- cations in scientific computing

Michael Bader. 2012.Space-filling curves: an introduction with appli- cations in scientific computing. Vol. 9. Springer Science & Business Media

2012
[4]

Bampis, Zhi Li, Joel Sole, and Alan C

Li-Heng Chen, Christos G. Bampis, Zhi Li, Joel Sole, and Alan C. Bovik
[5]

doi:10.1109/TIP.2020.3043127

Perceptual Video Quality Prediction Emphasizing Chroma Dis- tortions.IEEE Transactions on Image Processing30 (2021), 1408–1422. doi:10.1109/TIP.2020.3043127

work page doi:10.1109/tip.2020.3043127 2021
[6]

NVIDIA Corporation. 2025. NVIDIA nvCOMP - A CUDA library for Fast Lossless Compression. https://developer.nvidia.com/nvcomp

2025
[7]

Ricardo L De Queiroz and Philip A Chou. 2016. Compression of 3D point clouds using a region-adaptive hierarchical transform.IEEE Transactions on Image Processing25, 8 (2016), 3947–3956

2016
[8]

E d’Eon, B Harrison, T Myers, and PA Chou. 2017. 8i Vox- elized Full Bodies—A Voxelized Point Cloud Dataset, document WG11M40059/WG1M74006.ISO/IEC JTC1/SC29 Joint WG11/WG1 (MPEG/JPEG), Geneva, Switzerland(2017)

2017
[9]

FFmpeg Developers. 2026. FFmpeg. http://ffmpeg.org/

2026
[10]

Jiaye Fu, Qiankun Gao, Chengxiang Wen, Yanmin Wu, Siwei Ma, Jiaqi Zhang, and Jian Zhang. 2025. ReCon-GS: Continuum-Preserved Gauss- ian Streaming for Fast and Compact Reconstruction of Dynamic Scenes. InAdvances in Neural Information Processing Systems (NeurIPS)

2025
[11]

Sharath Girish, Tianye Li, Amrita Mazumdar, Abhinav Shrivastava, david luebke, and Shalini De Mello. 2024. QUEEN: QUantized Efficient ENcoding for Streaming Free-viewpoint Videos. InThe Thirty-eighth Annual Conference on Neural Information Processing Systems. https: //openreview.net/forum?id=7xhwE7VH4S

2024
[12]

Yongjie Guan, Xueyu Hou, Nan Wu, Bo Han, and Tao Han. 2023. MetaStream: Live Volumetric Content Capture, Creation, Delivery, and Rendering in Real Time. InProceedings of the 29th Annual Interna- tional Conference on Mobile Computing and Networking. Association for Computing Machinery, New York, NY, USA, Article 29, 15 pages. https://doi.org/10.1145/357036...

work page doi:10.1145/3570361.3592530 2023
[13]

Bo Han, Yu Liu, and Feng Qian. 2020. ViVo: Visibility-Aware Mobile Volumetric Video Streaming. InProceedings of the 26th Annual Inter- national Conference on Mobile Computing and Networking(London, United Kingdom)(MobiCom ’20). Association for Computing Machin- ery, New York, NY, USA, Article 11, 13 pages. doi:10.1145/3372224. 3380888

work page doi:10.1145/3372224 2020
[14]

Yuning Huang, Jiahao Pang, Fengqing Zhu, and Dong Tian. 2026. EntropyGS: An Efficient Entropy Coding on 3D Gaussian Splatting. In ICASSP 2026 - 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

2026
[15]

Netflix Inc. 2015. Per-title Encode Optimization. https://netflixtechblog. com/per-title-encode-optimization-7e99442b62a2

2015
[16]

Tao Jin, Mallesham Dasa, Connor Smith, Kittipat Apicharttrisorn, Srini- vasan Seshan, and Anthony Rowe. 2024. MeshReduce: Scalable and Bandwidth Efficient 3D Scene Capture. In2024 IEEE Conference Virtual Reality and 3D User Interfaces (VR). 20–30. doi:10.1109/VR58804.2024. 00026

work page doi:10.1109/vr58804.2024 2024
[17]

Hanbyul Joo, Tomas Simon, Xulong Li, Hao Liu, Lei Tan, Lin Gui, Sean Banerjee, Timothy Scott Godisart, Bart Nabbe, Iain Matthews, Takeo Kanade, Shohei Nobuhara, and Yaser Sheikh. 2017. Panoptic Studio: A Massively Multiview System for Social Interaction Capture.IEEE Transactions on Pattern Analysis and Machine Intelligence(2017)

2017
[18]

Tero Karras. 2012. Maximizing parallelism in the construction of BVHs, octrees, and k-d trees. InProceedings of the Fourth ACM SIGGRAPH / Eurographics Conference on High-Performance Graphics(Paris, France) (EGGH-HPG’12). Eurographics Association, Goslar, DEU, 33–37

2012
[19]

Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, and George Drettakis. 2023. 3D Gaussian Splatting for Real-Time Radiance Field Rendering.ACM Transactions on Graphics42, 4 (July 2023). https: //repo-sam.inria.fr/fungraph/3d-gaussian-splatting/

2023
[20]

Gunjoong Kim, Seonghoon Park, Jeho Lee, Chanyoung Jung, Hyung- chol Jun, and Hojung Cha. 2025. Vega: Fully Immersive Mobile Vol- umetric Video Streaming with 3D Gaussian Splatting. InProceedings of the 31st Annual International Conference on Mobile Computing and Networking(Kerry Hotel, Hong Kong, Hong Kong, China)(ACM MO- BICOM ’25). Association for Compu...

work page doi:10.1145/3680207.3765267 2025
[21]

Naimin Koh, Pradeep Kumar Jayaraman, and Jianmin Zheng. 2020. Par- allel Point Cloud Compression Using Truncated Octree. In2020 Inter- national Conference on Cyberworlds (CW). 1–8. doi:10.1109/CW49994. 2020.00009

work page doi:10.1109/cw49994 2020
[22]

Joo Chan Lee, Daniel Rho, Xiangyu Sun, Jong Hwan Ko, and Eunbyung Park. 2024. Compact 3D Gaussian Splatting for Static and Dynamic Radiance Fields. arXiv:2408.03822 [cs.CV] https://arxiv.org/abs/2408. 03822

arXiv 2024
[23]

Kyungjin Lee, Juheon Yi, Youngki Lee, Sunghyun Choi, and Young Min Kim. 2020. GROOT: a real-time streaming system of high-fidelity vol- umetric videos. InProceedings of the 26th Annual International Confer- ence on Mobile Computing and Networking(London, United Kingdom) (MobiCom ’20). Association for Computing Machinery, New York, NY, USA, Article 57, 14 ...

work page doi:10.1145/3372224.3419214 2020
[24]

Xiangrui Liu, Xinju Wu, Shiqi Wang, Zhu Li, and Sam Kwong. 2025. CompGS++: Compressed Gaussian Splatting for Static and Dynamic Scene Representation. arXiv:2504.13022 [cs.GR] https://arxiv.org/abs/ 2504.13022

arXiv 2025
[25]

Tao Lu, Mulin Yu, Linning Xu, Yuanbo Xiangli, Limin Wang, Dahua Lin, and Bo Dai. 2024. Scaffold-gs: Structured 3d gaussians for view- adaptive rendering. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition. 20654–20664

2024
[26]

H.S. Malvar. 2006. Adaptive run-length/Golomb-Rice encoding of quantized generalized Gaussian sources with unknown statistics. In Data Compression Conference (DCC’06). 23–32. doi:10.1109/DCC.2006.5

work page doi:10.1109/dcc.2006.5 2006
[27]

Donald Meagher. 1982. Geometric modeling using octree encoding. Computer Graphics and Image Processing19, 2 (1982), 129–147. doi:10. 1016/0146-664X(82)90104-6

1982
[28]

MPEG. 2023. GPCC - mpeg-pcc-tmc13. https://github.com/ MPEGGroup/mpeg-pcc-tmc13

2023
[29]

K L Navaneet, Kossar Pourahmadi Meibodi, Soroush Abbasi Kooh- payegani, and Hamed Pirsiavash. 2024. CompGS: Smaller and Faster Gaussian Splatting with Vector Quantization. InComputer Vision 13 - ECCV 2024: 18th European Conference, Milan, Italy, September 29- October 4, 2024, Proceedings, Part XXXII(Milan, Italy). Springer-Verlag, Berlin, Heidelberg, 330–...

work page doi:10.1007/978-3-031-73411-3_19 2024
[30]

NVIDIA. 2024. NVIDIA Video Codec SDK. https://developer.nvidia. com/video-codec-sdk

2024
[31]

NVIDIA. 2024. Video Encode and Decode GPU Support Ma- trix. https://developer.nvidia.com/video-encode-and-decode-gpu- support-matrix-new

2024
[32]

NVIDIA Corporation. 2024. NVIDIA Jetson AGX Orin Devel- oper Kit. https://www.nvidia.com/en-us/autonomous-machines/ embedded-systems/jetson-orin/ Accessed: 2024-05-20

2024
[33]

2019.PyTorch: an imperative style, high-performance deep learning library

Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Brad- bury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas Köpf, Edward Yang, Zach DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner, Lu Fang, Junjie Bai, and Soumith Chintala. 2019.PyTorch: an imperative style, high-p...

2019
[34]

Chou, Ricardo L

Eduardo Pavez, Philip A. Chou, Ricardo L. de Queiroz, and Antonio Or- tega. 2018. Dynamic polygon clouds: representation and compression for VR/AR.APSIPA Transactions on Signal and Information Processing 7 (2018), e15. doi:10.1017/ATSIP.2018.15

work page doi:10.1017/atsip.2018.15 2018
[35]

Rao and N

K. Rao and N. Ahmed. 1976. Orthogonal transforms for digital signal processing. InICASSP ’76. IEEE International Conference on Acoustics, Speech, and Signal Processing, Vol. 1. 136–140. doi:10.1109/ICASSP. 1976.1170121

work page doi:10.1109/icassp 1976
[36]

Chou, Robert A

Sebastian Schwarz, Marius Preda, Vittorio Baroncini, Madhukar Buda- gavi, Pablo Cesar, Philip A. Chou, Robert A. Cohen, Maja Krivokuća, Sébastien Lasserre, Zhu Li, Joan Llach, Khaled Mammou, Rufael Mekuria, Ohji Nakagami, Ernestasia Siahaan, Ali Tabatabai, Alexis M. Tourapis, and Vladyslav Zakharchenko. 2019. Emerging MPEG Standards for Point Cloud Compre...

work page doi:10.1109/jetcas.2018.2885981 2019
[37]

Anil Shanbhag, Samuel Madden, and Xiangyao Yu. 2020. A Study of the Fundamental Performance Characteristics of GPUs and CPUs for Database Analytics. InProceedings of the 2020 ACM SIGMOD In- ternational Conference on Management of Data(Portland, OR, USA) (SIGMOD ’20). Association for Computing Machinery, New York, NY, USA, 1617–1632. doi:10.1145/3318464.3380595

work page doi:10.1145/3318464.3380595 2020
[38]

Yuang Shi. 2025. 3D Gaussian-based Immersive Media Streaming in Networked Extended Reality. InProceedings of the 16th ACM Multi- media Systems Conference(Stellenbosch, South Africa)(MMSys ’25). Association for Computing Machinery, New York, NY, USA, 356–360. doi:10.1145/3712676.3719673

work page doi:10.1145/3712676.3719673 2025
[39]

Sullivan and Shijun Sun

Gary J. Sullivan and Shijun Sun. 2005. On dead-zone plus uniform threshold scalar quantization. InVisual Communications and Image Processing 2005, Shipeng Li, Fernando Pereira, Heung-Yeung Shum, and Andrew G. Tescher (Eds.), Vol. 5960. International Society for Optics and Photonics, SPIE, 596033. doi:10.1117/12.631550

work page doi:10.1117/12.631550 2005
[40]

Yuan-Chun Sun, Yuang Shi, Cheng-Tse Lee, Mufeng Zhu, Wei Tsang Ooi, Yao Liu, Chun-Ying Huang, and Cheng-Hsin Hsu. 2025. LTS: A DASH Streaming System for Dynamic Multi-Layer 3D Gaussian Splatting Scenes. InProceedings of the 16th ACM Multimedia Systems Conference(Stellenbosch, South Africa)(MMSys ’25). Association for Computing Machinery, New York, NY, USA...

arXiv 2025
[41]

Yuan-Chun Sun, Yuang Shi, Wei Tsang Ooi, Chun-Ying Huang, and Cheng-Hsin Hsu. 2024. Multi-frame Bitrate Allocation of Dynamic 3D Gaussian Splatting Streaming Over Dynamic Networks. InProceedings of the 2024 SIGCOMM Workshop on Emerging Multimedia Systems(Syd- ney, NSW, Australia)(EMS ’24). Association for Computing Machinery, New York, NY, USA, 1–7. doi:1...

work page doi:10.1145/3672196.3673394 2024
[42]

Sampath, and George Biros

Hari Sundar, Rahul S. Sampath, and George Biros. 2008. Bottom- Up Construction and 2:1 Balance Refinement of Linear Octrees in Parallel.SIAM Journal on Scientific Computing30, 5 (2008), 2675–2708. arXiv:https://doi.org/10.1137/070681727 doi:10.1137/070681727

work page doi:10.1137/070681727 2008
[43]

Sridhara, Eduardo Pavez, Antonio Or- tega, and Cheng Chang

Chenjunjie Wang, Shashank N. Sridhara, Eduardo Pavez, Antonio Or- tega, and Cheng Chang. 2025. Adaptive Voxelization for Transform Coding of 3D Gaussian Splatting Data. In2025 IEEE International Con- ference on Image Processing (ICIP). 2414–2419. doi:10.1109/ICIP55913. 2025.11084522

work page doi:10.1109/icip55913 2025
[44]

Kangli Wang, Shihao Li, Qianxi Yi, and Wei Gao. 2025. A Novel Benchmark and Dataset for Efficient 3D Gaussian Splatting with Gaussian Point Cloud Compression. arXiv:2505.18197 [cs.GR] https: //arxiv.org/abs/2505.18197

arXiv 2025
[45]

Penghao Wang, Zhirui Zhang, Liao Wang, Kaixin Yao, Siyuan Xie, Jingyi Yu, Minye Wu, and Lan Xu. 2024. Vˆ 3: Viewing Volumet- ric Videos on Mobiles via Streamable 2D Dynamic Gaussians.ACM Transactions on Graphics (TOG)43, 6 (2024), 1–13

2024
[46]

Guanjun Wu, Taoran Yi, Jiemin Fang, Lingxi Xie, Xiaopeng Zhang, Wei Wei, Wenyu Liu, Qi Tian, and Xinggang Wang. 2024. 4D Gaussian Splatting for Real-Time Dynamic Scene Rendering. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 20310–20320

2024
[47]

Shuzhao Xie, Weixiang Zhang, Chen Tang, Yunpeng Bai, Rongwei Lu, Shijia Ge, and Zhi Wang. 2024. MesonGS: Post-training Compression of 3D Gaussians via Efficient Attribute Transformation. InEuropean Conference on Computer Vision. Springer

2024
[48]

Kai Zhang, Kaibo Wang, Yuan Yuan, Lei Guo, Rubao Lee, and Xiaodong Zhang. 2015. Mega-KV: a case for GPUs to maximize the throughput of in-memory key-value stores.Proc. VLDB Endow.8, 11 (July 2015), 1226–1237. doi:10.14778/2809974.2809984

work page doi:10.14778/2809974.2809984 2015
[49]

Shunyuan Zheng, Boyao Zhou, Ruizhi Shao, Boning Liu, Shengping Zhang, Liqiang Nie, and Yebin Liu. 2024. GPS-Gaussian: Generalizable Pixel-wise 3D Gaussian Splatting for Real-time Human Novel View Synthesis. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

2024
[50]

Zihan Zheng, Zhenlong Wu, Houqiang Zhong, Yuan Tian, Ning Cao, Lan Xu, Jiangchao Yao, Xiaoyun Zhang, Qiang Hu, and Wenjun Zhang
[51]

arXiv:2509.17513 [cs.CV] https://arxiv.org/abs/2509.17513

4DGCPro: Efficient Hierarchical 4D Gaussian Compression for Progressive Volumetric Video Streaming. arXiv:2509.17513 [cs.CV] https://arxiv.org/abs/2509.17513

arXiv
[52]

Boyao Zhou, Shunyuan Zheng, Hanzhang Tu, Ruizhi Shao, Boning Liu, Shengping Zhang, Liqiang Nie, and Yebin Liu. 2024. GPS-Gaussian+: Generalizable Pixel-wise 3D Gaussian Splatting for Real-Time Human- Scene Rendering from Sparse Views.arXiv preprint arXiv:2411.11363 (2024)

arXiv 2024
[53]

Kun Zhou, Minmin Gong, Xin Huang, and Baining Guo. 2011. Data- Parallel Octrees for Surface Reconstruction.IEEE Transactions on Visualization and Computer Graphics17, 5 (2011), 669–681. doi:10. 1109/TVCG.2010.75 14 A Coding Latency In Table A.1, we report the actual encoding and decoding latencies for all sequences in the HiFi4G and N3DV datasets. B GS-NF...

2011

[1] [1]

[n. d.]. Draco 3D Compression. https://github.com/google/draco

[2] [2]

[n. d.]. Point Cloud Library. https://pointclouds.org/

[3] [3]

2012.Space-filling curves: an introduction with appli- cations in scientific computing

Michael Bader. 2012.Space-filling curves: an introduction with appli- cations in scientific computing. Vol. 9. Springer Science & Business Media

2012

[4] [4]

Bampis, Zhi Li, Joel Sole, and Alan C

Li-Heng Chen, Christos G. Bampis, Zhi Li, Joel Sole, and Alan C. Bovik

[5] [5]

doi:10.1109/TIP.2020.3043127

Perceptual Video Quality Prediction Emphasizing Chroma Dis- tortions.IEEE Transactions on Image Processing30 (2021), 1408–1422. doi:10.1109/TIP.2020.3043127

work page doi:10.1109/tip.2020.3043127 2021

[6] [6]

NVIDIA Corporation. 2025. NVIDIA nvCOMP - A CUDA library for Fast Lossless Compression. https://developer.nvidia.com/nvcomp

2025

[7] [7]

Ricardo L De Queiroz and Philip A Chou. 2016. Compression of 3D point clouds using a region-adaptive hierarchical transform.IEEE Transactions on Image Processing25, 8 (2016), 3947–3956

2016

[8] [8]

E d’Eon, B Harrison, T Myers, and PA Chou. 2017. 8i Vox- elized Full Bodies—A Voxelized Point Cloud Dataset, document WG11M40059/WG1M74006.ISO/IEC JTC1/SC29 Joint WG11/WG1 (MPEG/JPEG), Geneva, Switzerland(2017)

2017

[9] [9]

FFmpeg Developers. 2026. FFmpeg. http://ffmpeg.org/

2026

[10] [10]

Jiaye Fu, Qiankun Gao, Chengxiang Wen, Yanmin Wu, Siwei Ma, Jiaqi Zhang, and Jian Zhang. 2025. ReCon-GS: Continuum-Preserved Gauss- ian Streaming for Fast and Compact Reconstruction of Dynamic Scenes. InAdvances in Neural Information Processing Systems (NeurIPS)

2025

[11] [11]

Sharath Girish, Tianye Li, Amrita Mazumdar, Abhinav Shrivastava, david luebke, and Shalini De Mello. 2024. QUEEN: QUantized Efficient ENcoding for Streaming Free-viewpoint Videos. InThe Thirty-eighth Annual Conference on Neural Information Processing Systems. https: //openreview.net/forum?id=7xhwE7VH4S

2024

[12] [12]

Yongjie Guan, Xueyu Hou, Nan Wu, Bo Han, and Tao Han. 2023. MetaStream: Live Volumetric Content Capture, Creation, Delivery, and Rendering in Real Time. InProceedings of the 29th Annual Interna- tional Conference on Mobile Computing and Networking. Association for Computing Machinery, New York, NY, USA, Article 29, 15 pages. https://doi.org/10.1145/357036...

work page doi:10.1145/3570361.3592530 2023

[13] [13]

Bo Han, Yu Liu, and Feng Qian. 2020. ViVo: Visibility-Aware Mobile Volumetric Video Streaming. InProceedings of the 26th Annual Inter- national Conference on Mobile Computing and Networking(London, United Kingdom)(MobiCom ’20). Association for Computing Machin- ery, New York, NY, USA, Article 11, 13 pages. doi:10.1145/3372224. 3380888

work page doi:10.1145/3372224 2020

[14] [14]

Yuning Huang, Jiahao Pang, Fengqing Zhu, and Dong Tian. 2026. EntropyGS: An Efficient Entropy Coding on 3D Gaussian Splatting. In ICASSP 2026 - 2026 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

2026

[15] [15]

Netflix Inc. 2015. Per-title Encode Optimization. https://netflixtechblog. com/per-title-encode-optimization-7e99442b62a2

2015

[16] [16]

Tao Jin, Mallesham Dasa, Connor Smith, Kittipat Apicharttrisorn, Srini- vasan Seshan, and Anthony Rowe. 2024. MeshReduce: Scalable and Bandwidth Efficient 3D Scene Capture. In2024 IEEE Conference Virtual Reality and 3D User Interfaces (VR). 20–30. doi:10.1109/VR58804.2024. 00026

work page doi:10.1109/vr58804.2024 2024

[17] [17]

Hanbyul Joo, Tomas Simon, Xulong Li, Hao Liu, Lei Tan, Lin Gui, Sean Banerjee, Timothy Scott Godisart, Bart Nabbe, Iain Matthews, Takeo Kanade, Shohei Nobuhara, and Yaser Sheikh. 2017. Panoptic Studio: A Massively Multiview System for Social Interaction Capture.IEEE Transactions on Pattern Analysis and Machine Intelligence(2017)

2017

[18] [18]

Tero Karras. 2012. Maximizing parallelism in the construction of BVHs, octrees, and k-d trees. InProceedings of the Fourth ACM SIGGRAPH / Eurographics Conference on High-Performance Graphics(Paris, France) (EGGH-HPG’12). Eurographics Association, Goslar, DEU, 33–37

2012

[19] [19]

Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, and George Drettakis. 2023. 3D Gaussian Splatting for Real-Time Radiance Field Rendering.ACM Transactions on Graphics42, 4 (July 2023). https: //repo-sam.inria.fr/fungraph/3d-gaussian-splatting/

2023

[20] [20]

Gunjoong Kim, Seonghoon Park, Jeho Lee, Chanyoung Jung, Hyung- chol Jun, and Hojung Cha. 2025. Vega: Fully Immersive Mobile Vol- umetric Video Streaming with 3D Gaussian Splatting. InProceedings of the 31st Annual International Conference on Mobile Computing and Networking(Kerry Hotel, Hong Kong, Hong Kong, China)(ACM MO- BICOM ’25). Association for Compu...

work page doi:10.1145/3680207.3765267 2025

[21] [21]

Naimin Koh, Pradeep Kumar Jayaraman, and Jianmin Zheng. 2020. Par- allel Point Cloud Compression Using Truncated Octree. In2020 Inter- national Conference on Cyberworlds (CW). 1–8. doi:10.1109/CW49994. 2020.00009

work page doi:10.1109/cw49994 2020

[22] [22]

Joo Chan Lee, Daniel Rho, Xiangyu Sun, Jong Hwan Ko, and Eunbyung Park. 2024. Compact 3D Gaussian Splatting for Static and Dynamic Radiance Fields. arXiv:2408.03822 [cs.CV] https://arxiv.org/abs/2408. 03822

arXiv 2024

[23] [23]

Kyungjin Lee, Juheon Yi, Youngki Lee, Sunghyun Choi, and Young Min Kim. 2020. GROOT: a real-time streaming system of high-fidelity vol- umetric videos. InProceedings of the 26th Annual International Confer- ence on Mobile Computing and Networking(London, United Kingdom) (MobiCom ’20). Association for Computing Machinery, New York, NY, USA, Article 57, 14 ...

work page doi:10.1145/3372224.3419214 2020

[24] [24]

Xiangrui Liu, Xinju Wu, Shiqi Wang, Zhu Li, and Sam Kwong. 2025. CompGS++: Compressed Gaussian Splatting for Static and Dynamic Scene Representation. arXiv:2504.13022 [cs.GR] https://arxiv.org/abs/ 2504.13022

arXiv 2025

[25] [25]

Tao Lu, Mulin Yu, Linning Xu, Yuanbo Xiangli, Limin Wang, Dahua Lin, and Bo Dai. 2024. Scaffold-gs: Structured 3d gaussians for view- adaptive rendering. InProceedings of the IEEE/CVF Conference on Com- puter Vision and Pattern Recognition. 20654–20664

2024

[26] [26]

H.S. Malvar. 2006. Adaptive run-length/Golomb-Rice encoding of quantized generalized Gaussian sources with unknown statistics. In Data Compression Conference (DCC’06). 23–32. doi:10.1109/DCC.2006.5

work page doi:10.1109/dcc.2006.5 2006

[27] [27]

Donald Meagher. 1982. Geometric modeling using octree encoding. Computer Graphics and Image Processing19, 2 (1982), 129–147. doi:10. 1016/0146-664X(82)90104-6

1982

[28] [28]

MPEG. 2023. GPCC - mpeg-pcc-tmc13. https://github.com/ MPEGGroup/mpeg-pcc-tmc13

2023

[29] [29]

K L Navaneet, Kossar Pourahmadi Meibodi, Soroush Abbasi Kooh- payegani, and Hamed Pirsiavash. 2024. CompGS: Smaller and Faster Gaussian Splatting with Vector Quantization. InComputer Vision 13 - ECCV 2024: 18th European Conference, Milan, Italy, September 29- October 4, 2024, Proceedings, Part XXXII(Milan, Italy). Springer-Verlag, Berlin, Heidelberg, 330–...

work page doi:10.1007/978-3-031-73411-3_19 2024

[30] [30]

NVIDIA. 2024. NVIDIA Video Codec SDK. https://developer.nvidia. com/video-codec-sdk

2024

[31] [31]

NVIDIA. 2024. Video Encode and Decode GPU Support Ma- trix. https://developer.nvidia.com/video-encode-and-decode-gpu- support-matrix-new

2024

[32] [32]

NVIDIA Corporation. 2024. NVIDIA Jetson AGX Orin Devel- oper Kit. https://www.nvidia.com/en-us/autonomous-machines/ embedded-systems/jetson-orin/ Accessed: 2024-05-20

2024

[33] [33]

2019.PyTorch: an imperative style, high-performance deep learning library

Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Brad- bury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas Köpf, Edward Yang, Zach DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner, Lu Fang, Junjie Bai, and Soumith Chintala. 2019.PyTorch: an imperative style, high-p...

2019

[34] [34]

Chou, Ricardo L

Eduardo Pavez, Philip A. Chou, Ricardo L. de Queiroz, and Antonio Or- tega. 2018. Dynamic polygon clouds: representation and compression for VR/AR.APSIPA Transactions on Signal and Information Processing 7 (2018), e15. doi:10.1017/ATSIP.2018.15

work page doi:10.1017/atsip.2018.15 2018

[35] [35]

Rao and N

K. Rao and N. Ahmed. 1976. Orthogonal transforms for digital signal processing. InICASSP ’76. IEEE International Conference on Acoustics, Speech, and Signal Processing, Vol. 1. 136–140. doi:10.1109/ICASSP. 1976.1170121

work page doi:10.1109/icassp 1976

[36] [36]

Chou, Robert A

Sebastian Schwarz, Marius Preda, Vittorio Baroncini, Madhukar Buda- gavi, Pablo Cesar, Philip A. Chou, Robert A. Cohen, Maja Krivokuća, Sébastien Lasserre, Zhu Li, Joan Llach, Khaled Mammou, Rufael Mekuria, Ohji Nakagami, Ernestasia Siahaan, Ali Tabatabai, Alexis M. Tourapis, and Vladyslav Zakharchenko. 2019. Emerging MPEG Standards for Point Cloud Compre...

work page doi:10.1109/jetcas.2018.2885981 2019

[37] [37]

Anil Shanbhag, Samuel Madden, and Xiangyao Yu. 2020. A Study of the Fundamental Performance Characteristics of GPUs and CPUs for Database Analytics. InProceedings of the 2020 ACM SIGMOD In- ternational Conference on Management of Data(Portland, OR, USA) (SIGMOD ’20). Association for Computing Machinery, New York, NY, USA, 1617–1632. doi:10.1145/3318464.3380595

work page doi:10.1145/3318464.3380595 2020

[38] [38]

Yuang Shi. 2025. 3D Gaussian-based Immersive Media Streaming in Networked Extended Reality. InProceedings of the 16th ACM Multi- media Systems Conference(Stellenbosch, South Africa)(MMSys ’25). Association for Computing Machinery, New York, NY, USA, 356–360. doi:10.1145/3712676.3719673

work page doi:10.1145/3712676.3719673 2025

[39] [39]

Sullivan and Shijun Sun

Gary J. Sullivan and Shijun Sun. 2005. On dead-zone plus uniform threshold scalar quantization. InVisual Communications and Image Processing 2005, Shipeng Li, Fernando Pereira, Heung-Yeung Shum, and Andrew G. Tescher (Eds.), Vol. 5960. International Society for Optics and Photonics, SPIE, 596033. doi:10.1117/12.631550

work page doi:10.1117/12.631550 2005

[40] [40]

Yuan-Chun Sun, Yuang Shi, Cheng-Tse Lee, Mufeng Zhu, Wei Tsang Ooi, Yao Liu, Chun-Ying Huang, and Cheng-Hsin Hsu. 2025. LTS: A DASH Streaming System for Dynamic Multi-Layer 3D Gaussian Splatting Scenes. InProceedings of the 16th ACM Multimedia Systems Conference(Stellenbosch, South Africa)(MMSys ’25). Association for Computing Machinery, New York, NY, USA...

arXiv 2025

[41] [41]

Yuan-Chun Sun, Yuang Shi, Wei Tsang Ooi, Chun-Ying Huang, and Cheng-Hsin Hsu. 2024. Multi-frame Bitrate Allocation of Dynamic 3D Gaussian Splatting Streaming Over Dynamic Networks. InProceedings of the 2024 SIGCOMM Workshop on Emerging Multimedia Systems(Syd- ney, NSW, Australia)(EMS ’24). Association for Computing Machinery, New York, NY, USA, 1–7. doi:1...

work page doi:10.1145/3672196.3673394 2024

[42] [42]

Sampath, and George Biros

Hari Sundar, Rahul S. Sampath, and George Biros. 2008. Bottom- Up Construction and 2:1 Balance Refinement of Linear Octrees in Parallel.SIAM Journal on Scientific Computing30, 5 (2008), 2675–2708. arXiv:https://doi.org/10.1137/070681727 doi:10.1137/070681727

work page doi:10.1137/070681727 2008

[43] [43]

Sridhara, Eduardo Pavez, Antonio Or- tega, and Cheng Chang

Chenjunjie Wang, Shashank N. Sridhara, Eduardo Pavez, Antonio Or- tega, and Cheng Chang. 2025. Adaptive Voxelization for Transform Coding of 3D Gaussian Splatting Data. In2025 IEEE International Con- ference on Image Processing (ICIP). 2414–2419. doi:10.1109/ICIP55913. 2025.11084522

work page doi:10.1109/icip55913 2025

[44] [44]

Kangli Wang, Shihao Li, Qianxi Yi, and Wei Gao. 2025. A Novel Benchmark and Dataset for Efficient 3D Gaussian Splatting with Gaussian Point Cloud Compression. arXiv:2505.18197 [cs.GR] https: //arxiv.org/abs/2505.18197

arXiv 2025

[45] [45]

Penghao Wang, Zhirui Zhang, Liao Wang, Kaixin Yao, Siyuan Xie, Jingyi Yu, Minye Wu, and Lan Xu. 2024. Vˆ 3: Viewing Volumet- ric Videos on Mobiles via Streamable 2D Dynamic Gaussians.ACM Transactions on Graphics (TOG)43, 6 (2024), 1–13

2024

[46] [46]

Guanjun Wu, Taoran Yi, Jiemin Fang, Lingxi Xie, Xiaopeng Zhang, Wei Wei, Wenyu Liu, Qi Tian, and Xinggang Wang. 2024. 4D Gaussian Splatting for Real-Time Dynamic Scene Rendering. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). 20310–20320

2024

[47] [47]

Shuzhao Xie, Weixiang Zhang, Chen Tang, Yunpeng Bai, Rongwei Lu, Shijia Ge, and Zhi Wang. 2024. MesonGS: Post-training Compression of 3D Gaussians via Efficient Attribute Transformation. InEuropean Conference on Computer Vision. Springer

2024

[48] [48]

Kai Zhang, Kaibo Wang, Yuan Yuan, Lei Guo, Rubao Lee, and Xiaodong Zhang. 2015. Mega-KV: a case for GPUs to maximize the throughput of in-memory key-value stores.Proc. VLDB Endow.8, 11 (July 2015), 1226–1237. doi:10.14778/2809974.2809984

work page doi:10.14778/2809974.2809984 2015

[49] [49]

Shunyuan Zheng, Boyao Zhou, Ruizhi Shao, Boning Liu, Shengping Zhang, Liqiang Nie, and Yebin Liu. 2024. GPS-Gaussian: Generalizable Pixel-wise 3D Gaussian Splatting for Real-time Human Novel View Synthesis. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

2024

[50] [50]

Zihan Zheng, Zhenlong Wu, Houqiang Zhong, Yuan Tian, Ning Cao, Lan Xu, Jiangchao Yao, Xiaoyun Zhang, Qiang Hu, and Wenjun Zhang

[51] [51]

arXiv:2509.17513 [cs.CV] https://arxiv.org/abs/2509.17513

4DGCPro: Efficient Hierarchical 4D Gaussian Compression for Progressive Volumetric Video Streaming. arXiv:2509.17513 [cs.CV] https://arxiv.org/abs/2509.17513

arXiv

[52] [52]

Boyao Zhou, Shunyuan Zheng, Hanzhang Tu, Ruizhi Shao, Boning Liu, Shengping Zhang, Liqiang Nie, and Yebin Liu. 2024. GPS-Gaussian+: Generalizable Pixel-wise 3D Gaussian Splatting for Real-Time Human- Scene Rendering from Sparse Views.arXiv preprint arXiv:2411.11363 (2024)

arXiv 2024

[53] [53]

Kun Zhou, Minmin Gong, Xin Huang, and Baining Guo. 2011. Data- Parallel Octrees for Surface Reconstruction.IEEE Transactions on Visualization and Computer Graphics17, 5 (2011), 669–681. doi:10. 1109/TVCG.2010.75 14 A Coding Latency In Table A.1, we report the actual encoding and decoding latencies for all sequences in the HiFi4G and N3DV datasets. B GS-NF...

2011