Pith. sign in

REVIEW 5 major objections 6 minor 63 references

Efficient Geometry Compression and Communication for 3D Gaussian Splatting Point Clouds

T0 review · 5 major / 6 minor · reviewed 2026-08-05 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Adding AVS PCRM lossless geometry compression to the i3DV Gaussian-splatting platform cuts total bitrate by 10-25% on four dynamic-scene test sets while PSNR, SSIM, and LPIPS remain exactly unchanged.

desk verdict Plausible incremental savings from wiring AVS PCRM into i3DV, but the lossless round trip is under-specified and the paper's numbers don't always add up. read the letter →

arxiv 2509.02232 v1 pith:W7I2IFFY submitted 2025-09-02 cs.MM

Liang Xie , Yanting Li , Luyang Tang , Wei Gao This is my paper
classification cs.MM
keywords 3DGaussianSplattingPointCloudCompressionAVSPCRMi3DVVolumetricVideoLosslessGeometryCodingMortonCodeRate-Distortion
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to show that the geometry part of a 3D Gaussian volumetric-video stream can be handed to an existing point-cloud codec — AVS PCRM in lossless mode — without hurting rendering quality. Plugging this codec into the i3DV platform replaces uncompressed anchor geometry with a compressed bitstream, and the reported result is a 10-25% total bitrate reduction on four 250-frame test sequences, with geometry-only rates down 55-69%. Because the geometry coding is lossless and attributes are aligned by Morton order, all three quality metrics (PSNR, SSIM, LPIPS) come out exactly equal to the baseline. The practical stake is that dynamic 3D scenes become cheaper to store and stream, including within a 40 Mbps bandwidth budget, without retraining the Gaussian model.

What carries the argument

Morton-code spatial ordering: each 3D position is mapped to a one-dimensional space-filling order, and both geometry and per-Gaussian attributes are sorted by that order before lossless AVS PCRM encoding and rebuilt in the same order after decoding. This alignment is the mechanism that lets a generic point-cloud geometry codec replace the platform's geometry storage while keeping every attribute attached to the correct Gaussian point.

What would settle it

Take one frame's anchor data from i3DV, run it through the described save_ply, avs_encoder, avs_decoder, load_ply cycle, and compare the decoded point coordinates and point count byte-for-byte with the original numpy arrays; any coordinate change, duplicate-point merge, or count difference means the Morton alignment cannot restore exact correspondence. A complementary check is to render the decoded anchors and look for any pixel difference, since exact PSNR/SSIM/LPIPS equality to the baseline is a direct consequence of losslessness.

Watch

Extended reading notes

Core claim

On the i3DV platform, the paper replaces the original uncompressed geometry encoding of anchor-based Gaussian point clouds with AVS PCRM v11 lossless point-cloud geometry coding. Anchor data is exported to PLY, compressed by the avs_encoder, decoded back by avs_decoder, and reloaded; attributes continue to be coded by i3DV's binary hash table that caches inter-frame Gaussian transformations. To prevent the decoder from reordering points and detaching attributes from their Gaussians, geometry is sorted by Morton code before encoding and all attributes are rearranged by the same Morton indices, restoring one-to-one correspondence after decoding. On Dance_Dunhuang_Pair_1080, Show_Groups_4K, VRU

Load-bearing premise

The whole scheme assumes that PLY export followed by AVS PCRM lossless encoding and decoding returns exactly the same point set — same coordinates, same count, same order after Morton sorting — so that attributes reattached by index are still attached to the correct Gaussian; the paper does not specify the numeric precision of the PLY export or the lossless configuration.

Editorial extensions

If this is right

  • Total bitrate for the four tested dynamic Gaussian sequences drops 10-25% with identical PSNR, SSIM, and LPIPS values.
  • Geometry-only bitrate drops 55-69%, so the saving grows when a scene has more geometric complexity relative to attribute data.
  • The same rendering and rate-distortion behavior of i3DV is preserved, including high-fidelity transmission within a 40 Mbps constraint.
  • Encoding and decoding time stays about the same as the baseline, so the gain is obtained without a noticeable compute penalty.
  • The compression is a drop-in replacement for the geometry path and requires no retraining or modification of the Gaussian model.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the reported quality metrics are exactly identical to the baseline, the geometry path is effectively bit-exact in practice; any future lossy geometry variant would need a different attribute-alignment strategy.
  • The Morton-order alignment trick should transfer to other Gaussian-splatting pipelines that store geometry and attributes as separate arrays and can export PLY, not just i3DV and AVS PCRM.
  • The 10-25% total savings are likely proportional to geometry's share of the stream; as SH order or attribute dimensions grow, attribute coding rather than geometry will become the bottleneck.
  • A natural extension is to evaluate lossy AVS PCRM configurations, which would trade the exact-quality guarantee for larger bitrate savings and require a way to propagate geometry distortion into attribute alignment.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 6 minor

Summary. The paper proposes replacing the geometry coding step in the i3DV dynamic 3D Gaussian platform with the AVS PCRM point-cloud codec in lossless mode. The proposed pipeline converts i3DV anchor data to PLY, encodes the geometry with AVS PCRM, decodes it, and re-aligns geometry and attributes using Morton-code sorting. The reported experiments on four 250-frame sequences show identical PSNR/SSIM/LPIPS values between the original i3DV and the proposed integration, together with 10.14%-25.64% total-size reductions and 55-69% geometry-size reductions. The introduction claims a 40% storage reduction, while the abstract and conclusion state 10%-25%.

Significance. If the central claim is correct, the contribution is practically useful: a drop-in replacement of one geometry codec that reduces total bitrate by 10-25% while keeping rendering quality bit-exactly unchanged is an attractive engineering result for volumetric video streaming. The paper has the merit of reporting results on standard multi-view datasets over 250 frames, including encode/decode wall-clock times, and it builds on publicly available components (i3DV and AVS PCRM), which makes reproduction in principle possible. However, the significance is conditional on two currently unverified premises: (i) the PLY/AVS-PCRM round trip is truly lossless at the numerical precision of the Gaussian anchor data, and (ii) the reported 'Size' comparison uses identical, fully specified bit-accounting rules for baseline and proposed methods. The paper provides no RD curves, no error bars, and only one operating point per sequence, so the 'superior rate-distortion tradeoff' claim in Section 5 is not directly evidenced.

major comments (5)
  1. [Sec. 3.1-3.2, Table 1] The load-bearing losslessness claim is not established. The paper never specifies the PLY export precision, the coordinate scaling/quantization used by the AVS PCRM lossless mode, or the exact decoder configuration. Moreover, Sec. 3.2 states that 'lossless compression' introduces an inconsistency in positional relationships before/after encoding, which is contradictory: a lossless codec may permute point order but must preserve point positions. If positions actually change, then the exact equality of PSNR, SSIM, and LPIPS in Table 1 across 250 frames and four sequences is not a logical consequence of losslessness; if positions do not change, the Morton alignment is a permutation and should be described and verified as such. Please report the numerical precision of every conversion step and demonstrate bit-exact equality of the decoded geometry, or otherwise recompute the quality metrics
  2. [Sec. 4, Table 1] The baseline geometry storage format is never defined. It is unclear whether 'XYZ Size (MB)' for i3DV2.0 is raw float32, quantized integer coordinates, or an already compressed geometry bitstream, and whether 'Size (MB)' includes the full 59-channel Gaussian attributes or only a subset. The 10-25% total-size savings cannot be reproduced or even interpreted without knowing exactly which bytes are being replaced by the AVS PCRM bitstream, whether the Morton-order side information is included in the reported size, and how the PLY header/attribute encoding is accounted for. Please provide a precise byte-level breakdown of the baseline and proposed storage formats.
  3. [Sec. 1 vs. Abstract/Conclusion] The introduction states the solution 'achieves a 40% storage reduction,' whereas the abstract and Section 5 claim 10%-25%. Table 1 reports per-sequence total reductions of 10.14%, 13.40%, 25.64%, and 24.23%. The 40% figure is not derivable from the reported data and must be corrected or tied to a different experimental condition.
  4. [Sec. 4.2, Table 1] There are multiple internal numerical inconsistencies. The text says Dance_Dunhuang 'Size is 0.834 MB for both methods with a compression rate of 10.14%,' but the table lists 0.8324 MB and 0.748 MB. Show_Groups_4K is described as 0.971 MB while the table lists 0.9171 and 0.7942 MB; VRU_gz is described as 2.394 MB while the table lists 2.9555 and 2.2394 MB. The statement that encoding time is 1.582s/1.589s 'for both' is also inconsistent with the table's proposed-method values. These discrepancies undermine confidence in the table's accuracy and must be reconciled.
  5. [Sec. 4.2 and Sec. 5] The paper evaluates exactly one operating point per sequence and reports no rate-distortion curves or error bars. Because the proposed geometry coding is lossless, the rendered quality is expected to be identical to the baseline; in that setting the contribution is a pure bitrate reduction at fixed distortion. The phrase 'superior rate-distortion tradeoff' in Section 5 is therefore not supported unless the authors vary the geometry quantization (or another coding parameter) and show RD behavior across multiple points. At minimum, state explicitly that the comparison is a fixed-quality, single-point bitrate comparison.
minor comments (6)
  1. [Abstract] The first sentence is grammatically incomplete: 'Storage and transmission challenges in dynamic 3D scene representation based on the i3DV platform, With increasing scene complexity...' should be split and rewritten.
  2. [Sec. 2.2] The text contains a stray LaTeX citation remnant, '@inproceedingsSPCGC, G-PCC suffers...', which should be removed.
  3. [Sec. 4.2] The phrase 'AVS PCRM has a smaller file size on VRU g_gz' contains a typo ('g_gz'). Also, the LPIPS discussion claims a 'slight improvement in the VRU_gz version' although the reported LPIPS values are exactly equal; clarify whether this refers to a different metric or is a wording error.
  4. [Sec. 4.1] The dataset description calls all four sequences 'point cloud datasets,' but the experiments are 3D Gaussian rendering sequences derived from multi-view video. Please clarify the relationship between the raw multi-view data, the reconstructed Gaussian anchor sets, and the PLY point clouds actually encoded.
  5. [References] Reference [56] (Yang and Mandt, conditional diffusion image compression) is cited for 'compression rate [56]' in Sec. 4.2, which appears irrelevant to point-cloud/Gaussian bitrate. Please replace with an appropriate citation or remove.
  6. [Throughout] The paper uses 'i3DV2.0 + AVS PCRM [19]' to denote the proposed method, but AVS PCRM is a codec, not a competing algorithm; the comparison labels in Table 1 and the text should distinguish clearly between the i3DV baseline and the i3DV-plus-PCRM integrated system.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the central claim is an empirical bitrate comparison between two external codecs; self-citations are contextual, not load-bearing reductions.

full rationale

The paper's contribution is an experimental integration of two existing systems: i3DV (prior work by co-authors) and AVS PCRM (a standard reference codec whose development involves co-authors). The claimed 10%-25% total bitrate savings (Abstract, Sec 5, Table 1) are measured file sizes of the two pipelines on four external datasets. The identical PSNR/SSIM/LPIPS values are presented as a sanity check of lossless geometry coding, not as a fitted or predicted quantity. None of the paper's equations, definitions, or cited theorems force the savings: the savings come from AVS PCRM's geometry bitstream size versus i3DV's original geometry representation, an external empirical comparison. The self-citations [19,37,38,41] support the existence and capabilities of the two platforms, but the comparative measurements are not derived from those citations. The closest definitional element is that 'lossless' geometry coding makes the quality metrics identical by construction; however, the paper does not present that equality as an independent prediction, and the actual contribution (bitrate reduction) is not circular. I also considered the robustness gap flagged by a skeptical reader: Sec 3.1-3.2 never specifies PLY export precision or the AVS PCRM lossless configuration, so if the round trip quantizes coordinates, the exact identical metrics would not follow. That is a correctness/verifiability concern, not a circular reasoning concern, and it does not involve a fitted parameter being renamed as a prediction. No load-bearing argument reduces to its own inputs via self-citation or definitional construction, so the circularity score is low.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The method rests on two external systems (i3DV and AVS PCRM) and on the correctness of the Morton alignment assumption. No new entities, forces, or fitted models are introduced; the only unstated free choice is the AVS PCRM encoder configuration.

free parameters (1)
  • AVS PCRM encoder operating point = unspecified
    The compression results depend on AVS PCRM's quantization/configuration settings, which are never reported. Without this, the claimed 10-25% savings are not anchored to a defined bitrate or quality operating point.
assumptions (3)
  • domain assumption Morton-order reordering preserves the one-to-one correspondence between decoded geometry and attributes after lossless compression (Sec 3.2).
    The whole alignment scheme assumes that sorting both geometry and attributes by Morton code yields the same permutation on both sides, and that AVS PCRM's lossless mode changes only point order, not point coordinates.
  • domain assumption The four test sequences (Dance_Dunhuang_Pair_1080, Show_Groups_4K, VRU_dg4, VRU_gz) are representative of 'universal test sets' (Sec 4.1).
    The abstract and conclusion generalize to universal test sets from only two basketball sequences and two human sequences, all processed at one operating point.
  • domain assumption AVS PCRM v11 lossless geometry coding exactly preserves the PLY-converted positions (Sec 3.1).
    The method never states the numeric precision of the PLY export or shows that no quantization occurs; the exactly-identical PSNR/SSIM/LPIPS values depend on this being true.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Efficient Geometry Compression and Communication for 3D Gaussian Splatting Point Clouds." pith.science (2026). https://pith.science/paper/W7I2IFFY

@misc{pith2026250902232,
  author       = {Pith},
  title        = {Pith review of: Efficient Geometry Compression and Communication for 3D Gaussian Splatting Point Clouds},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/W7I2IFFY}},
  note         = {Machine review of arXiv:2509.02232}
}
read the original abstract

Storage and transmission challenges in dynamic 3D scene representation based on the i3DV platform, With increasing scene complexity, the explosive growth of 3D Gaussian data volume causes excessive storage space occupancy. To address this issue, we propose adopting the AVS PCRM reference software for efficient compression of Gaussian point cloud geometry data. The strategy deeply integrates the advanced encoding capabilities of AVS PCRM into the i3DV platform, forming technical complementarity with the original rate-distortion optimization mechanism based on binary hash tables. On one hand, the hash table efficiently caches inter-frame Gaussian point transformation relationships, which allows for high-fidelity transmission within a 40 Mbps bandwidth constraint. On the other hand, AVS PCRM performs precise compression on geometry data. Experimental results demonstrate that the joint framework maintains the advantages of fast rendering and high-quality synthesis in 3D Gaussian technology while achieving significant 10\%-25\% bitrate savings on universal test sets. It provides a superior rate-distortion tradeoff solution for the storage, transmission, and interaction of 3D volumetric video.

Figures

Figures reproduced from arXiv: 2509.02232 by the authors.

Figure 1
Figure 1. The geometry and attribute data processing pipeline in i3DV [ [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 3
Figure 3. The reconstruction results of different algo [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. The reconstruction results of different algo [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

63 extracted references · 57 canonical work pages

  1. [1]

    Milena T Bagdasarian, Paul Knoll, Y Li, Florian Barthel, Anna Hils- mann, Peter Eisert, and Wieland Morgenstern. 2024. 3dgs.zip: A survey on 3d gaussian splatting compression methods. In Computer Graphics Forum. Wiley Online Library, e70078

    work page 2024

  2. [2]

    Yihang Chen, Mengyao Li, Qianyi Wu, Weiyao Lin, Mehrtash Harandi, and Jianfei Cai. 2025. Pcgs: Progressive compression of 3d gaussian splatting. arXiv preprint arXiv:2503.08511 (2025)

    arXiv 2025

  3. [3]

    Yihang Chen, Qianyi Wu, Weiyao Lin, Mehrtash Harandi, and Jianfei Cai. 2024. Hac: Hash-grid assisted context for 3d gaussian splatting compression. In European Conference on Computer Vision . Springer, 422–438

    work page 2024

  4. [4]

    Zicong Chen, Zhenghao Chen, Wei Jiang, Wei Wang, Lei Liu, and Dong Xu. 2025. 4DGS-CC: A Contextual Coding Framework for 4D Gaussian Splatting Data Compression. arXiv preprint arXiv:2504.18925 (2025)

    work page Pith review arXiv 2025

  5. [5]

    Ricardo de Queiroz, Andre Souto, Victor Figueiredo, and Philip Chou

    work page

  6. [6]

    Ricardo L De Queiroz and Philip A Chou. 2016. Compression of 3D point clouds using a region-adaptive hierarchical transform. IEEE Transactions on Image Processing 25, 8 (2016), 3947–3956

    work page 2016

  7. [7]

    Lue Fan, Yuxue Yang, Minxing Li, Hongsheng Li, and Zhaoxiang Zhang

    work page

  8. [8]

    Zhiwen Fan, Kevin Wang, Kairun Wen, Zehao Zhu, Dejia Xu, Zhangyang Wang, et al. 2024. Lightgaussian: Unbounded 3d gaussian compression with 15x reduction and 200+ fps. Advances in neural information processing systems 37 (2024), 140138–140158

    work page 2024

Show all 63 references
  1. [9]

    Yang Fu, Sifei Liu, Amey Kulkarni, Jan Kautz, Alexei A Efros, and Xiaolong Wang. 2024. Colmap-free 3d gaussian splatting. In IEEE/CVF Conference on Computer Vision and Pattern Recognition . 20796–20805

    2024

  2. [10]

    Pan Gao, Shengzhou Luo, and Manoranjan Paul. 2022. Rate-distortion modeling for bit rate constrained point cloud compression. IEEE Transactions on Circuits and Systems for Video Technology 33, 5 (2022), 2424–2438

    2022

  3. [11]

    Wei Gao, Liang Xie, Songlin Fan, and Ge Li. 2025. Deep Learning-Based Point Cloud Compression: An In-Depth Survey and Benchmark. IEEE Transactions on Pattern Analysis and Machine Intelligence (2025)

    2025

  4. [12]

    Wei Gao, Hua Ye, Ge Li, Huiming Zheng, Yuyang Wu, and Liang Xie. 2022. OpenPointCloud: An Open-Source Algorithm Library of Deep Learning Based Point Cloud Compression. In ACM International Conference on Multimedia. 7347–7350

    2022

  5. [13]

    Danillo Graziosi, Ohji Nakagami, Satoru Kuma, Alexandre Zaghetto, Teruhiko Suzuki, and Ali Tabatabai. 2020. An overview of ongoing point cloud compression standardization activities: Video-based (V- PCC) and geometry-based (G-PCC). APSIPA Transactions on Signal and Information...

    2020

  6. [14]

    Shuai Gu, Junhui Hou, Huanqiang Zeng, and Hui Yuan. 2020. 3D point cloud attribute compression via graph prediction. IEEE Signal Processing Letters 27 (2020), 176–180

    2020

  7. [15]

    Tian Guo, Hui Yuan, Lu Wang, and Tingting Wang. 2023. Rate- distortion optimized quantization for geometry-based point cloud compression. Journal of Electronic Imaging32, 1 (2023), 013047–013047

    2023

  8. [16]

    He Huang, Wenjie Huang, Qi Yang, Yiling Xu, and Zhu Li. 2025. A hierarchical compression technique for 3d gaussian splatting compres- sion. In IEEE International Conference on Acoustics, Speech and Signal Processing. IEEE, 1–5

    2025

  9. [17]

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, and George Drettakis. 2023. 3d gaussian splatting for real-time radiance field rendering. ACM Trans. Graph. 42, 4 (2023), 139–1

    2023

  10. [18]

    Maja Krivokuća, Ehsan Miandji, Christine Guillemot, and Philip A Chou. 2021. Compression of plenoptic point cloud attributes using 6-D point clouds and 6-D transforms. IEEE Transactions on Multimedia 25 (2021), 593–607

    2021

  11. [19]

    Ge Li, Wei Gao, and Wen Gao. 2024. AVS point cloud compression standard. In Point Cloud Compression: Technologies and Standardization. Springer, 167–197

    2024

  12. [20]

    Ge Li, Wei Gao, and Wen Gao. 2024. MPEG geometry-based point cloud compression (G-PCC) standard. In Point Cloud Compression: Technologies and Standardization. Springer, 135–165

    2024

  13. [21]

    Ge Li, Wei Gao, and Wen Gao. 2024. Point Cloud Compression: Tech- nologies and Standardization. Springer Nature

    2024

  14. [22]

    Li Li, Zhu Li, and Shan Liu. 2020. Occupancy-map-based rate distortion optimization and partition for video-based point cloud compression. IEEE Transactions on Circuits and Systems for Video Technology 31, 1 (2020), 326–338

    2020

  15. [23]

    Li Li, Zhu Li, Shan Liu, and Houqiang Li. 2020. Rate control for video- based point cloud compression. IEEE Transactions on Image Processing 29 (2020), 6237–6250

    2020

  16. [24]

    Sicheng Li, Chengzhen Wu, Hao Li, Xiang Gao, Yiyi Liao, and Lu Yu

  17. [25]

    Lei Liu, Zhenghao Chen, Wei Jiang, Wei Wang, and Dong Xu. 2024. Hemgs: A hybrid entropy model for 3d gaussian splatting data com- pression. arXiv preprint arXiv:2411.18473 (2024)

    2024 arXiv

  18. [26]

    Qi Liu, Hui Yuan, Raouf Hamzaoui, Honglei Su, Junhui Hou, and Huan Yang. 2021. Reduced reference perceptual quality model with application to rate control for video-based point cloud compression. IEEE Transactions on Image Processing 30 (2021), 6623–6636

    2021

  19. [27]

    Xueyuan Liu, Zhuoran Song, Hao Chen, and Xing Li. 2024. MoC: A Morton-Code-Based Fine-Grained Quantization for Accelerating Point Cloud Neural Networks. In ACM/IEEE Design Automation Conference. 1–6

    2024

  20. [28]

    Xiangrui Liu, Xinju Wu, Shiqi Wang, Zhu Li, and Sam Kwong. 2025. CompGS++: Compressed Gaussian Splatting for Static and Dynamic Scene Representation. arXiv preprint arXiv:2504.13022 (2025)

    2025 arXiv

  21. [29]

    Xiangrui Liu, Xinju Wu, Pingping Zhang, Shiqi Wang, Zhu Li, and Sam Kwong. 2024. Compgs: Efficient 3d scene representation via compressed gaussian splatting. In ACM International Conference on Multimedia. 2936–2944

    2024

  22. [30]

    Simone Milani, Enrico Polo, and Simone Limuti. 2020. A transform coding strategy for dynamic point clouds. IEEE Transactions on Image Processing 29 (2020), 8213–8225

    2020

  23. [31]

    Simon Niedermayr, Josef Stumpfegger, and Rüdiger Westermann. 2024. Compressed 3d gaussian splatting for accelerated novel view synthesis. In IEEE/CVF Conference on Computer Vision and Pattern Recognition . 10349–10358

    2024

  24. [32]

    Evaristo Ramalho, Eduardo Peixoto, and Edil Medeiros. 2021. Sil- houette 4D with context selection: Lossless geometry compression of dynamic point clouds. IEEE Signal Processing Letters 28 (2021), 1660–1664

    2021

  25. [33]

    Sungryeul Rhyu, Junsik Kim, Jiheon Im, and Kyuheon Kim. 2020. Con- textual homogeneity-based patch decomposition method for higher point cloud compression. IEEE Access 8 (2020), 207805–207812

    2020

  26. [34]

    Havard Rue and Leonhard Held. 2005. Gaussian Markov random fields: theory and applications. Chapman and Hall/CRC

    2005

  27. [35]

    Stefania Sardellitti, Sergio Barbarossa, and Paolo Di Lorenzo. 2017. On the graph Fourier transform for directed graphs. IEEE Journal of Selected Topics in Signal Processing 11, 6 (2017), 796–811

    2017

  28. [36]

    Johannes L Schonberger and Jan-Michael Frahm. 2016. Structure-from- motion revisited. In IEEE conference on computer vision and pattern recognition. 4104–4113

    2016

  29. [37]

    Luyang Tang, Jiayu Yang, Rui Peng, and Yongqi Zhai. 2025. Compress- ing streamable free-viewpoint videos to 0.1 mb per frame. In AAAI Conference on Artificial Intelligence, Vol. 39. 7257–7265

    2025

  30. [38]

    August, 2024

    Luyang Tang, Jiayu Yang, Rui Peng, and Yongqi Zhai. August, 2024. i3DV: Intelligent 3D Volumetric Video Encoding Platform.A VS, M8499, Online (August, 2024)

    2024

  31. [39]

    Renwei Tu, Gangyi Jiang, Mei Yu, Ting Luo, Zongju Peng, and Fen Chen. 2022. V-PCC projection based blind point cloud quality assess- ment for compression distortion. IEEE Transactions on Emerging Topics in Computational Intelligence 7, 2 (2022), 462–473

    2022

  32. [40]

    VRU. 2024. VRU-sequence. https://anonymous.4open.science/r/VRU- Sequence/

    2024

  33. [41]

    Yang Wang, Wei Gao, and Xingming Mu. 2023. Rate control optimiza- tion for joint geometry and attribute coding of lidar point clouds. In IEEE International Conference on Visual Communications and Image Processing. IEEE, 1–5

    2023

  34. [42]

    Yufei Wang, Zhihao Li, Lanqing Guo, Wenhan Yang, Alex Kot, and Bihan Wen. 2024. Contextgs: Compact 3d gaussian splatting with anchor level context model. Advances in neural information processing systems 37 (2024), 51532–51551

    2024

  35. [43]

    Yuxuan Wei, Zehan Wang, Tian Guo, Hao Liu, Liquan Shen, and Hui Yuan. 2025. High efficiency Wiener filter-based point cloud quality enhancement for MPEG G-PCC. IEEE Transactions on Circuits and Systems for Video Technology (2025). ACM MOBICOM ’25, November 4–8, 2025, Hong Kong...

    2025

  36. [44]

    Yuyang Wu, Liang Xie, Shangkun Sun, Wei Gao, and Yiqiang Yan. 2024. Adaptive Intra Period Size for Deep Learning-based Screen Content Video Coding. In IEEE International Conference on Multimedia and Expo Workshops. IEEE, 1–6

    2024

  37. [45]

    Liang Xie and Songlin Fan. 2025. A Learning-based Multi-Frame Visual Feature Framework for Real-Time Driver Fatigue Detection. In Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies . 61–69

    2025

  38. [46]

    Liang Xie and Wei Gao. 2024. LearningPCC: A PyTorch Library for Learning-Based Point Cloud Compression. In ACM International Con- ference on Multimedia. 11234–11238

    2024

  39. [47]

    Liang Xie and Wei Gao. 2024. PCHMVision: An Open-Source Library of Point Cloud Compression for Human and Machine Vision. In ACM International Conference on Multimedia . 11239–11243

    2024

  40. [48]

    Liang Xie, Wei Gao, Songlin Fan, and Zhaojian Yao. 2024. PDNet: Parallel Dual-branch Network for Point Cloud Geometry Compression and Analysis. In Data Compression Conference. 596–596

    2024

  41. [49]

    Liang Xie, Wei Gao, and Huiming Zheng. 2022. End-to-End Point Cloud Geometry Compression and Analysis with Sparse Tensor. In International Workshop on Advances in Point Cloud Compression, Pro- cessing and Analysis. 27–32

    2022

  42. [50]

    Liang Xie, Wei Gao, Huiming Zheng, and Ge Li. 2024. ROI-Guided Point Cloud Geometry Compression Towards Human and Machine Vision. In ACM International Conference on Multimedia . 3741–3750

    2024

  43. [51]

    Liang Xie, Wei Gao, Huiming Zheng, and Hua Ye. 2024. Semantic- Aware Visual Decomposition for Point Cloud Geometry Compression. In Data Compression Conference. 595–595

    2024

  44. [52]

    Liang Xie, MengHao Hu, and XinBei Bai. 2022. Online Improved Vehicle Tracking Accuracy via Unsupervised Route Generation. In IEEE 34th International Conference on Tools with Artificial Intelligence . IEEE, 788–792

    2022

  45. [53]

    Liang Xie, MengHao Hu, and XinBei Bai. 2022. Towards Hardware- Friendly and Robust Facial Landmark Detection Method. In Interna- tional Conference on Neural Information Processing . Springer, 432–444

    2022

  46. [54]

    Liang Xie, Xingming Mu, Ge Li, Wei Gao, et al. 2024. PKU-DPCC: A New Dataset for Dynamic Point Cloud Compression. APSIPA Transac- tions on Signal and Information Processing 13, 6 (2024)

    2024

  47. [55]

    Yiqun Xu, Wei Hu, Shanshe Wang, Xinfeng Zhang, Shiqi Wang, Siwei Ma, Zongming Guo, and Wen Gao. 2020. Predictive generalized graph Fourier transform for attribute compression of dynamic point clouds. IEEE Transactions on Circuits and Systems for Video Technology 31, 5 (2020), ...

    2020

  48. [56]

    Ruihan Yang and Stephan Mandt. 2023. Lossy image compression with conditional diffusion models. Advances in Neural Information Processing Systems 36 (2023), 64971–64995

    2023

  49. [57]

    Richard Zhang, Phillip Isola, Alexei A Efros, Eli Shechtman, and Oliver Wang. 2018. The unreasonable effectiveness of deep features as a perceptual metric. In IEEE conference on computer vision and pattern recognition. 586–595

    2018

  50. [58]

    Xinjie Zhang, Xingtong Ge, Tongda Xu, Dailan He, Yan Wang, Hong- wei Qin, Guo Lu, Jing Geng, and Jun Zhang. 2024. Gaussianimage: 1000 fps image representation and compression by 2d gaussian splatting. In European Conference on Computer Vision . Springer, 327–345

    2024

  51. [59]

    Tiesong Zhao, Jiheng Wang, Zhou Wang, and Chang Wen Chen. 2015. SSIM-based coarse-grain scalable video coding. IEEE Transactions on Broadcasting 61, 2 (2015), 210–221

    2015

  52. [60]

    Xiaoyun Zheng, Liwei Liao, Xufeng Li, Jianbo Jiao, Rongjie Wang, Feng Gao, Shiqi Wang, and Ronggang Wang. 2024. Pku-dymvhumans: A multi-view video benchmark for high-fidelity dynamic human model- ing. In IEEE/CVF Conference on Computer Vision and Pattern Recognition. 22530–22540

    2024

  53. [2023]

    Authorea Preprints (2023)

    Set Partitioning in Hierarchical Trees for Point Cloud Attribute Compression. Authorea Preprints (2023)

    2023

  54. [2024]

    arXiv preprint arXiv:2406.07499 (2024)

    Trim 3d gaussian splatting for accurate geometry representation. arXiv preprint arXiv:2406.07499 (2024). Efficient Geometry Compression and Communication for 3D Gaussian Splatting Point Clouds ACM MOBICOM ’25, November 4–8, 2025, Hong Kong, China

    2024 arXiv

  55. [2025]

    arXiv preprint arXiv:2506.01822 (2025)

    GSCodec Studio: A Modular Framework for Gaussian Splat Compression. arXiv preprint arXiv:2506.01822 (2025)

    2025 arXiv

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.