pith. sign in

arxiv: 2509.13482 · v2 · pith:Y6TX5LGInew · submitted 2025-09-16 · 💻 cs.CV

Improving 3D Gaussian Splatting Compression by Scene-Adaptive Lattice Vector Quantization

Hao Xu , Xiaolin Wu , Xi Zhang This is my paper

Pith reviewed 2026-05-21 21:40 UTC · model grok-4.3

classification 💻 cs.CV
keywords 3D Gaussian Splattingcompressionlattice vector quantizationscene adaptiverate distortion optimizationneural renderingquantization
0
0 comments X

The pith

Replacing uniform scalar quantization with scene-adaptive lattice vector quantization improves rate-distortion performance in 3D Gaussian Splatting compression with minimal overhead.

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

The paper aims to show that more advanced quantizers can enhance 3DGS compression beyond the simple uniform scalar quantization currently used in anchor-based neural methods. By introducing scene-adaptive lattice vector quantization, it captures specific characteristics of each scene to achieve better efficiency. This matters because 3DGS models produce huge amounts of data, and better compression makes them more practical for storage, transmission, and real-time applications without significant quality loss. The approach balances the benefits of vector quantization with low complexity and allows one model to support multiple bit rates.

Core claim

The central discovery is that scene-adaptive lattice vector quantization (SALVQ), achieved by optimizing the lattice basis for each scene, can replace uniform scalar quantization in existing 3DGS compression pipelines. This leads to enhanced rate-distortion performance with minimal modifications and computational overhead. Additionally, scaling the lattice basis vectors allows dynamic adjustment of lattice density for different bit rate targets using a single model.

What carries the argument

Scene-adaptive lattice vector quantization (SALVQ), which optimizes the lattice basis per scene to better capture characteristics and uses scaling for multi-rate support.

If this is right

  • Existing anchor-based 3DGS compression architectures can integrate SALVQ seamlessly to boost their performance.
  • A single trained model can accommodate multiple compression levels by adjusting lattice density through scaling.
  • Training time and memory consumption decrease since separate models for different rates are no longer needed.
  • Overall data size for 3DGS representations reduces while maintaining photorealistic rendering quality.

Where Pith is reading between the lines

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

  • Applying similar adaptive quantization could benefit other neural rendering techniques like NeRF variants.
  • This flexibility might enable on-the-fly rate adaptation in streaming applications for 3D content.
  • Further optimization of the lattice basis could lead to even lower complexity implementations.

Load-bearing premise

Per-scene optimization of the lattice basis adds negligible extra overhead and computational cost while reliably improving rate-distortion efficiency over a fixed lattice.

What would settle it

Running compression experiments on multiple scenes and comparing the bit rate versus distortion curves of SALVQ against USQ; if no consistent improvement or high overhead is observed, the claim would be falsified.

Figures

Figures reproduced from arXiv: 2509.13482 by Hao Xu, Xiaolin Wu, Xi Zhang.

Figure 1
Figure 1. Figure 1: Illustrating the pipeline of anchor-based neural compression methods, using Hash-Assisted Context (HAC) [14] as an [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Illustration of the advantages of LVQ over USQ. For equal cell area, lattices whose Voronoi cells more closely [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: R–D curves of HAC [14], ContextGS [18] and HAC++ [17] under different quantizers. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Visual comparison of different quantizers on four scenes: ‘playroom’ (DeepBlending [60]), ‘flower’ and ‘stump’ (Mip [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Single-rate vs. variable-rate compression performance on three architectures and datasets. Here, ‘VBR’ indicates that [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Comparison of two quantizers in the PCGS [57] architecture [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
read the original abstract

3D Gaussian Splatting (3DGS) is rapidly gaining popularity for its photorealistic rendering quality and real-time performance, but it generates massive amounts of data. Hence compressing 3DGS data is necessary for the cost effectiveness of 3DGS models. Recently, several anchor-based neural compression methods have been proposed, achieving good 3DGS compression performance. However, they all rely on uniform scalar quantization (USQ) due to its simplicity. A tantalizing question is whether more sophisticated quantizers can improve the current 3DGS compression methods with very little extra overhead and minimal change to the system. The answer is yes by replacing USQ with lattice vector quantization (LVQ). To better capture scene-specific characteristics, we optimize the lattice basis for each scene, improving LVQ's adaptability and R-D efficiency. This scene-adaptive LVQ (SALVQ) strikes a balance between the R-D efficiency of vector quantization and the low complexity of USQ. SALVQ can be seamlessly integrated into existing 3DGS compression architectures, enhancing their R-D performance with minimal modifications and computational overhead. Moreover, by scaling the lattice basis vectors, SALVQ can dynamically adjust lattice density, enabling a single model to accommodate multiple bit rate targets. This flexibility eliminates the need to train separate models for different compression levels, significantly reducing training time and memory consumption.

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.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes replacing uniform scalar quantization (USQ) with scene-adaptive lattice vector quantization (SALVQ) inside existing anchor-based neural compressors for 3D Gaussian Splatting. The core idea is to optimize the lattice basis matrix per scene so that the quantizer better matches scene-specific statistics, thereby improving rate-distortion performance. The authors further claim that scaling the optimized basis vectors allows a single trained model to target multiple bit rates without retraining, and that the added per-scene optimization imposes only negligible encoding overhead and requires minimal architectural changes.

Significance. If the quantitative claims hold, the work would be useful because it offers a drop-in vector-quantization upgrade that preserves the simplicity of current 3DGS pipelines while potentially reducing the number of models that must be trained for different operating points. The explicit separation of a one-time scene-adaptive optimization from the subsequent encoding step is a pragmatic design choice that could be adopted by follow-up compression papers.

major comments (2)
  1. [Abstract, §4] Abstract and §4 (method): the central claim that per-scene lattice-basis optimization adds only 'minimal modifications and computational overhead' is load-bearing for both the seamless-integration and single-model multi-rate arguments, yet no timing tables, iteration counts, or FLOPs comparison against the USQ baseline are provided. Without these numbers it is impossible to verify that the overhead remains negligible when the optimization is performed at compression time for every new scene.
  2. [§5] §5 (experiments): the abstract asserts R-D gains, but the manuscript must report concrete BD-rate savings, PSNR/SSIM deltas, and error bars across multiple scenes and bit-rate points. If these results exist only in figures without tabulated values or statistical significance tests, the strength of the improvement claim cannot be assessed.
minor comments (2)
  1. [§3] Notation: the scaling factor applied to the lattice basis vectors for multi-rate support should be given an explicit symbol and its effect on the quantization step size should be derived in an equation.
  2. [Figure 3] Figure clarity: the lattice-point diagrams would benefit from an overlay of the optimized versus fixed basis vectors so readers can visually confirm the scene adaptation.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and will revise the manuscript accordingly to provide the requested quantitative details.

read point-by-point responses

  1. Referee: [Abstract, §4] Abstract and §4 (method): the central claim that per-scene lattice-basis optimization adds only 'minimal modifications and computational overhead' is load-bearing for both the seamless-integration and single-model multi-rate arguments, yet no timing tables, iteration counts, or FLOPs comparison against the USQ baseline are provided. Without these numbers it is impossible to verify that the overhead remains negligible when the optimization is performed at compression time for every new scene.

    Authors: We agree that explicit measurements are needed to substantiate the overhead claim. The per-scene optimization is a one-time gradient descent on the lattice basis matrix performed at encoding time, but we will add a new table in the revised manuscript reporting average wall-clock time, iteration counts until convergence, and total encoding time comparison versus the USQ baseline across the test scenes. This will allow direct verification that the added cost remains small. revision: yes

  2. Referee: [§5] §5 (experiments): the abstract asserts R-D gains, but the manuscript must report concrete BD-rate savings, PSNR/SSIM deltas, and error bars across multiple scenes and bit-rate points. If these results exist only in figures without tabulated values or statistical significance tests, the strength of the improvement claim cannot be assessed.

    Authors: We accept that tabulated numerical results would strengthen the experimental section. In the revision we will insert a table listing BD-rate savings relative to the USQ baseline, average PSNR and SSIM deltas, and standard deviations computed over all evaluated scenes for each operating point. Error bars will be added to the rate-distortion plots, and we will report the number of scenes and bit-rate points used to ensure the gains are statistically representative. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper introduces scene-adaptive lattice vector quantization (SALVQ) by optimizing lattice basis vectors per scene to replace uniform scalar quantization in existing 3DGS compression pipelines. This optimization produces an adapted basis that is then applied to vector quantization, yielding claimed R-D gains as an empirical outcome rather than a definitional identity. No steps reduce by construction to inputs via self-definition, fitted parameters renamed as predictions, or load-bearing self-citations. The derivation remains self-contained against external USQ baselines without invoking author-specific uniqueness theorems or smuggling ansatzes through citations. The per-scene adaptation is presented as an independent algorithmic choice whose overhead and performance benefits are asserted separately from any tautological reduction.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the premise that per-scene lattice optimization is both cheap and beneficial; this optimization step functions as a free parameter choice whose value is fitted to each scene.

free parameters (1)
  • scene-specific lattice basis
    Optimized per scene to capture characteristics; treated as a tunable parameter rather than derived from first principles.
axioms (1)
  • domain assumption Lattice vector quantization yields better rate-distortion trade-offs than uniform scalar quantization for the coefficient distributions arising in 3DGS compression.
    Invoked to justify replacement of USQ; no proof or citation of this property for the specific data is given in the abstract.

pith-pipeline@v0.9.0 · 5778 in / 1132 out tokens · 60954 ms · 2026-05-21T21:40:22.433070+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 1 Pith paper

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

  1. 3DTurboQuant: Training-Free Near-Optimal Quantization for 3D Reconstruction Models

    cs.CV 2026-04 conditional novelty 7.0

    3DTurboQuant achieves training-free near-optimal quantization for 3DGS and DUSt3R models via random rotations inducing Beta distributions, enabling precomputed Lloyd-Max quantizers that deliver 3.5x and 7.9x compressi...

Reference graph

Works this paper leans on

66 extracted references · 66 canonical work pages · cited by 1 Pith paper · 1 internal anchor

  1. [1]

    Prior-guided neural compression of visual data,

    H. Xu, “Prior-guided neural compression of visual data,” Ph.D. thesis, McMaster University, Hamilton, Canada, 2025. [Online]. Available: http://hdl.handle.net/11375/32258

    work page 2025

  2. [2]

    Nerf: Representing scenes as neural radiance fields for view synthesis,

    B. Mildenhall, P. P. Srinivasan, M. Tancik, J. T. Barron, R. Ramamoorthi, and R. Ng, “Nerf: Representing scenes as neural radiance fields for view synthesis,”Communications of the ACM, vol. 65, no. 1, pp. 99–106, 2021

    work page 2021

  3. [3]

    Neural volumes: Learning dynamic renderable volumes from images,

    S. Lombardi, T. Simon, J. Saragih, G. Schwartz, A. Lehrmann, and Y . Sheikh, “Neural volumes: Learning dynamic renderable volumes from images,”arXiv preprint arXiv:1906.07751, 2019

    work page arXiv 1906

  4. [4]

    Scene representation networks: Continuous 3d-structure-aware neural scene representations,

    V . Sitzmann, M. Zollh ¨ofer, and G. Wetzstein, “Scene representation networks: Continuous 3d-structure-aware neural scene representations,” Advances in neural information processing systems, vol. 32, 2019

    work page 2019

  5. [5]

    Local light field fusion: Practical view synthesis with prescriptive sampling guidelines,

    B. Mildenhall, P. P. Srinivasan, R. Ortiz-Cayon, N. K. Kalantari, R. Ra- mamoorthi, R. Ng, and A. Kar, “Local light field fusion: Practical view synthesis with prescriptive sampling guidelines,”ACM Transactions on Graphics (ToG), vol. 38, no. 4, pp. 1–14, 2019

    work page 2019

  6. [6]

    3d gaussian splatting for real-time radiance field rendering

    B. Kerbl, G. Kopanas, T. Leimk ¨uhler, and G. Drettakis, “3d gaussian splatting for real-time radiance field rendering.”ACM Trans. Graph., vol. 42, no. 4, pp. 139–1, 2023

    work page 2023

  7. [7]

    Reducing the memory footprint of 3d gaussian splatting,

    P. Papantonakis, G. Kopanas, B. Kerbl, A. Lanvin, and G. Drettakis, “Reducing the memory footprint of 3d gaussian splatting,”Proceedings of the ACM on Computer Graphics and Interactive Techniques, vol. 7, no. 1, pp. 1–17, 2024

    work page 2024

  8. [8]

    Compact 3d gaussian representation for radiance field,

    J. C. Lee, D. Rho, X. Sun, J. H. Ko, and E. Park, “Compact 3d gaussian representation for radiance field,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 21 719–21 728

    work page 2024

  9. [9]

    Lightgaus- sian: Unbounded 3d gaussian compression with 15x reduction and 200+ fps,

    Z. Fan, K. Wang, K. Wen, Z. Zhu, D. Xu, and Z. Wang, “Lightgaussian: Unbounded 3d gaussian compression with 15x reduction and 200+ fps,” arXiv preprint arXiv:2311.17245, 2023

    work page arXiv 2023

  10. [10]

    End-to- end rate-distortion optimized 3d gaussian representation,

    H. Wang, H. Zhu, T. He, R. Feng, J. Deng, J. Bian, and Z. Chen, “End-to- end rate-distortion optimized 3d gaussian representation,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 76–92

    work page 2024

  11. [11]

    Compgs: Smaller and faster gaussian splatting with vector quantization,

    K. Navaneet, K. Pourahmadi Meibodi, S. Abbasi Koohpayegani, and H. Pirsiavash, “Compgs: Smaller and faster gaussian splatting with vector quantization,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 330–349

    work page 2024

  12. [12]

    Compressed 3d gaussian splatting for accelerated novel view synthesis,

    S. Niedermayr, J. Stumpfegger, and R. Westermann, “Compressed 3d gaussian splatting for accelerated novel view synthesis,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion, 2024, pp. 10 349–10 358

    work page 2024

  13. [13]

    Scaffold-gs: Structured 3d gaussians for view-adaptive rendering,

    T. Lu, M. Yu, L. Xu, Y . Xiangli, L. Wang, D. Lin, and B. Dai, “Scaffold-gs: Structured 3d gaussians for view-adaptive rendering,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 20 654–20 664

    work page 2024

  14. [14]

    Hac: Hash-grid assisted context for 3d gaussian splatting compression,

    Y . Chen, Q. Wu, W. Lin, M. Harandi, and J. Cai, “Hac: Hash-grid assisted context for 3d gaussian splatting compression,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 422–438

    work page 2024

  15. [15]

    CAT-3DGS: A context-adaptive triplane approach to rate-distortion-optimized 3DGS compression,

    Y .-T. Zhan, C.-Y . Ho, H. Yang, Y .-H. Chen, J. C. Chiang, Y .-L. Liu, and W.-H. Peng, “CAT-3DGS: A context-adaptive triplane approach to rate-distortion-optimized 3DGS compression,” inThe Thirteenth International Conference on Learning Representations, 2025. [Online]. Available: https://openreview.net/forum?id=m3KuuE2ozw

    work page 2025

  16. [16]

    Hemgs: A hybrid entropy model for 3d gaussian splatting data compression,

    L. Liu, Z. Chen, and D. Xu, “Hemgs: A hybrid entropy model for 3d gaussian splatting data compression,”arXiv preprint arXiv:2411.18473, 2024. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 12

    work page arXiv 2024

  17. [17]

    Hac++: Towards 100x compression of 3d gaussian splatting,

    Y . Chen, Q. Wu, W. Lin, M. Harandi, and J. Cai, “Hac++: Towards 100x compression of 3d gaussian splatting,”IEEE Transactions on Pattern Analysis and Machine Intelligence, pp. 1–17, 2025

    work page 2025

  18. [18]

    Contextgs: Com- pact 3d gaussian splatting with anchor level context model,

    Y . Wang, Z. Li, L. Guo, W. Yang, A. Kot, and B. Wen, “Contextgs: Com- pact 3d gaussian splatting with anchor level context model,”Advances in neural information processing systems, vol. 37, pp. 51 532–51 551, 2024

    work page 2024

  19. [19]

    Eagles: Efficient accelerated 3d gaussians with lightweight encodings,

    S. Girish, K. Gupta, and A. Shrivastava, “Eagles: Efficient accelerated 3d gaussians with lightweight encodings,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 54–71

    work page 2024

  20. [20]

    Mesongs: Post-training compression of 3d gaussians via efficient attribute trans- formation,

    S. Xie, W. Zhang, C. Tang, Y . Bai, R. Lu, S. Ge, and Z. Wang, “Mesongs: Post-training compression of 3d gaussians via efficient attribute trans- formation,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 434–452

    work page 2024

  21. [21]

    Radsplat: Radiance field-informed gaussian splatting for robust real- time rendering with 900+ fps,

    M. Niemeyer, F. Manhardt, M.-J. Rakotosaona, M. Oechsle, D. Duck- worth, R. Gosula, K. Tateno, J. Bates, D. Kaeser, and F. Tombari, “Radsplat: Radiance field-informed gaussian splatting for robust real- time rendering with 900+ fps,”arXiv preprint arXiv:2403.13806, 2024

    work page arXiv 2024

  22. [22]

    Elmgs: Enhancing memory and computation scalability through compression for 3d gaussian splat- ting,

    M. S. Ali, S.-H. Bae, and E. Tartaglione, “Elmgs: Enhancing memory and computation scalability through compression for 3d gaussian splat- ting,”arXiv preprint arXiv:2410.23213, 2024

    work page arXiv 2024

  23. [23]

    Pup 3d-gs: Principled uncertainty pruning for 3d gaussian splatting,

    A. Hanson, A. Tu, V . Singla, M. Jayawardhana, M. Zwicker, and T. Goldstein, “Pup 3d-gs: Principled uncertainty pruning for 3d gaussian splatting,”arXiv preprint arXiv:2406.10219, 2024

    work page arXiv 2024

  24. [24]

    Efficientgs: Streamlining gaussian splatting for large-scale high-resolution scene representation,

    W. Liu, T. Guan, B. Zhu, L. Ju, Z. Song, D. Li, Y . Wang, and W. Yang, “Efficientgs: Streamlining gaussian splatting for large-scale high-resolution scene representation,”arXiv preprint arXiv:2404.12777, 2024

    work page arXiv 2024

  25. [25]

    Safeguardgs: 3d gaussian primitive pruning while avoiding catastrophic scene destruction,

    Y . Lee, Z. Zhang, and D. Fan, “Safeguardgs: 3d gaussian primitive pruning while avoiding catastrophic scene destruction,”arXiv preprint arXiv:2405.17793, 2024

    work page arXiv 2024

  26. [26]

    Mini-splatting: Representing scenes with a constrained number of gaussians,

    G. Fang and B. Wang, “Mini-splatting: Representing scenes with a constrained number of gaussians,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 165–181

    work page 2024

  27. [27]

    Point cloud attribute compression with graph transform,

    C. Zhang, D. Florencio, and C. Loop, “Point cloud attribute compression with graph transform,” in2014 IEEE International Conference on Image Processing (ICIP). IEEE, 2014, pp. 2066–2070

    work page 2014

  28. [28]

    Compression of 3d point clouds using a region-adaptive hierarchical transform,

    R. L. De Queiroz and P. A. Chou, “Compression of 3d point clouds using a region-adaptive hierarchical transform,”IEEE Transactions on Image Processing, vol. 25, no. 8, pp. 3947–3956, 2016

    work page 2016

  29. [29]

    3d point cloud attribute compression using geometry-guided sparse representation,

    S. Gu, J. Hou, H. Zeng, H. Yuan, and K.-K. Ma, “3d point cloud attribute compression using geometry-guided sparse representation,” IEEE Transactions on Image Processing, vol. 29, pp. 796–808, 2019

    work page 2019

  30. [30]

    A volumetric approach to point cloud compression—part i: Attribute compression,

    P. A. Chou, M. Koroteev, and M. Krivoku ´ca, “A volumetric approach to point cloud compression—part i: Attribute compression,”IEEE Trans- actions on Image Processing, vol. 29, pp. 2203–2216, 2019

    work page 2019

  31. [31]

    Attribute artifacts removal for geometry-based point cloud compression,

    X. Sheng, L. Li, D. Liu, and Z. Xiong, “Attribute artifacts removal for geometry-based point cloud compression,”IEEE Transactions on Image Processing, vol. 31, pp. 3399–3413, 2022

    work page 2022

  32. [32]

    Sparse tensor-based point cloud attribute com- pression,

    J. Wang and Z. Ma, “Sparse tensor-based point cloud attribute com- pression,” in2022 IEEE 5th International Conference on Multimedia Information Processing and Retrieval (MIPR). IEEE, 2022, pp. 59–64

    work page 2022

  33. [33]

    Density- preserving deep point cloud compression,

    Y . He, X. Ren, D. Tang, Y . Zhang, X. Xue, and Y . Fu, “Density- preserving deep point cloud compression,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 2333–2342

    work page 2022

  34. [34]

    Fast point cloud geometry compression with context-based residual coding and inr-based refinement,

    H. Xu, X. Zhang, and X. Wu, “Fast point cloud geometry compression with context-based residual coding and inr-based refinement,” inEuro- pean Conference on Computer Vision. Springer, 2024, pp. 270–288

    work page 2024

  35. [35]

    Channel-wise autoregressive entropy models for learned image compression,

    D. Minnen and S. Singh, “Channel-wise autoregressive entropy models for learned image compression,” in2020 IEEE International Conference on Image Processing (ICIP). IEEE, 2020, pp. 3339–3343

    work page 2020

  36. [36]

    Checkerboard context model for efficient learned image compression,

    D. He, Y . Zheng, B. Sun, Y . Wang, and H. Qin, “Checkerboard context model for efficient learned image compression,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 14 771–14 780

    work page 2021

  37. [37]

    Joint autoregressive and hierarchical priors for learned image compression,

    D. Minnen, J. Ball ´e, and G. D. Toderici, “Joint autoregressive and hierarchical priors for learned image compression,”Advances in neural information processing systems, vol. 31, 2018

    work page 2018

  38. [38]

    Lvqac: Lattice vector quantization coupled with spatially adaptive companding for efficient learned image compression,

    X. Zhang and X. Wu, “Lvqac: Lattice vector quantization coupled with spatially adaptive companding for efficient learned image compression,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 10 239–10 248

    work page 2023

  39. [39]

    Approaching rate-distortion limits in neural compression with lattice transform coding,

    E. Lei, H. Hassani, and S. S. Bidokhti, “Approaching rate-distortion limits in neural compression with lattice transform coding,” inThe Thirteenth International Conference on Learning Representations, 2025. [Online]. Available: https://openreview.net/forum?id=Tv36j85SqR

    work page 2025

  40. [40]

    Entropy relaxed lattice vector quantization for learned image compression,

    M. Cao, W. Dai, S. Li, H. Li, C. Li, J. Zou, and H. Xiong, “Entropy relaxed lattice vector quantization for learned image compression,” in 2024 Data Compression Conference (DCC). IEEE, 2024, pp. 548–548

    work page 2024

  41. [41]

    Learning optimal lattice vector quantizers for end-to-end neural image compression,

    X. Zhang and X. Wu, “Learning optimal lattice vector quantizers for end-to-end neural image compression,”Advances in Neural Information Processing Systems, vol. 37, pp. 106 497–106 518, 2024

    work page 2024

  42. [42]

    Multirate neural image compression with adaptive lattice vector quantization,

    H. Xu, X. Wu, and X. Zhang, “Multirate neural image compression with adaptive lattice vector quantization,” inProceedings of the Computer Vision and Pattern Recognition Conference (CVPR), June 2025, pp. 7633–7642

    work page 2025

  43. [43]

    Sayood,Introduction to data compression

    K. Sayood,Introduction to data compression. Morgan Kaufmann, 2017

    work page 2017

  44. [44]

    Fast quantizing and decoding and algorithms for lattice quantizers and codes,

    J. Conway and N. Sloane, “Fast quantizing and decoding and algorithms for lattice quantizers and codes,”IEEE Transactions on Information Theory, vol. 28, no. 2, pp. 227–232, 1982

    work page 1982

  45. [45]

    Gersho and R

    A. Gersho and R. M. Gray,V ector quantization and signal compression. Springer Science & Business Media, 2012, vol. 159

    work page 2012

  46. [46]

    J. H. Conway and N. J. A. Sloane,Sphere packings, lattices and groups. Springer Science & Business Media, 2013, vol. 290

    work page 2013

  47. [47]

    Soft-to-hard vector quantization for end- to-end learning compressible representations,

    E. Agustsson, F. Mentzer, M. Tschannen, L. Cavigelli, R. Timofte, L. Benini, and L. V . Gool, “Soft-to-hard vector quantization for end- to-end learning compressible representations,”Advances in neural in- formation processing systems, vol. 30, 2017

    work page 2017

  48. [48]

    Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation

    Y . Bengio, N. L ´eonard, and A. Courville, “Estimating or propagating gradients through stochastic neurons for conditional computation,”arXiv preprint arXiv:1308.3432, 2013

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  49. [49]

    Neural discrete representation learning,

    A. Van Den Oord, O. Vinyalset al., “Neural discrete representation learning,”Advances in neural information processing systems, vol. 30, 2017

    work page 2017

  50. [50]

    Sq-vae: Variational bayes on discrete representation with self-annealed stochastic quantization,

    Y . Takida, T. Shibuya, W. Liao, C.-H. Lai, J. Ohmura, T. Uesaka, N. Murata, S. Takahashi, T. Kumakura, and Y . Mitsufuji, “Sq-vae: Variational bayes on discrete representation with self-annealed stochastic quantization,”arXiv preprint arXiv:2205.07547, 2022

    work page arXiv 2022

  51. [51]

    Pytorch: An imperative style, high-performance deep learning library,

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antigaet al., “Pytorch: An imperative style, high-performance deep learning library,”Advances in neural information processing systems, vol. 32, 2019

    work page 2019

  52. [52]

    On lov ´asz’lattice reduction and the nearest lattice point problem,

    L. Babai, “On lov ´asz’lattice reduction and the nearest lattice point problem,”Combinatorica, vol. 6, pp. 1–13, 1986

    work page 1986

  53. [53]

    Variable bitrate image compression with quality scaling factors,

    T. Chen and Z. Ma, “Variable bitrate image compression with quality scaling factors,” inICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2020, pp. 2163–2167

    work page 2020

  54. [54]

    Asymmetric gained deep image compression with continuous rate adaptation,

    Z. Cui, J. Wang, S. Gao, T. Guo, Y . Feng, and B. Bai, “Asymmetric gained deep image compression with continuous rate adaptation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 10 532–10 541

    work page 2021

  55. [55]

    Qvrf: A quantization-error-aware variable rate framework for learned image com- pression,

    K. Tong, Y . Wu, Y . Li, K. Zhang, L. Zhang, and X. Jin, “Qvrf: A quantization-error-aware variable rate framework for learned image com- pression,” in2023 IEEE International Conference on Image Processing (ICIP). IEEE, 2023, pp. 1310–1314

    work page 2023

  56. [56]

    Variable-rate learned im- age compression with multi-objective optimization and quantization- reconstruction offsets,

    F. Kamisli, F. Racap ´e, and H. Choi, “Variable-rate learned im- age compression with multi-objective optimization and quantization- reconstruction offsets,” in2024 Data Compression Conference (DCC). IEEE, 2024, pp. 193–202

    work page 2024

  57. [57]

    Pcgs: Progressive compression of 3d gaussian splatting,

    Y . Chen, M. Li, Q. Wu, W. Lin, M. Harandi, and J. Cai, “Pcgs: Progressive compression of 3d gaussian splatting,”arXiv preprint arXiv:2503.08511, 2025

    work page arXiv 2025

  58. [58]

    Mip-nerf 360: Unbounded anti-aliased neural radiance fields,

    J. T. Barron, B. Mildenhall, D. Verbin, P. P. Srinivasan, and P. Hedman, “Mip-nerf 360: Unbounded anti-aliased neural radiance fields,” inPro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 5470–5479

    work page 2022

  59. [59]

    Tanks and temples: Benchmarking large-scale scene reconstruction,

    A. Knapitsch, J. Park, Q.-Y . Zhou, and V . Koltun, “Tanks and temples: Benchmarking large-scale scene reconstruction,”ACM Transactions on Graphics (ToG), vol. 36, no. 4, pp. 1–13, 2017

    work page 2017

  60. [60]

    Deep blending for free-viewpoint image-based rendering,

    P. Hedman, J. Philip, T. Price, J.-M. Frahm, G. Drettakis, and G. Bros- tow, “Deep blending for free-viewpoint image-based rendering,”ACM Transactions on Graphics (ToG), vol. 37, no. 6, pp. 1–15, 2018

    work page 2018

  61. [61]

    Image quality assessment: from error visibility to structural similarity,

    Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image quality assessment: from error visibility to structural similarity,”IEEE transactions on image processing, vol. 13, no. 4, pp. 600–612, 2004

    work page 2004

  62. [62]

    The unreasonable effectiveness of deep features as a perceptual metric,

    R. Zhang, P. Isola, A. A. Efros, E. Shechtman, and O. Wang, “The unreasonable effectiveness of deep features as a perceptual metric,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 586–595

    work page 2018

  63. [63]

    Calculation of average psnr differences between rd- curves,

    G. Bjontegaard, “Calculation of average psnr differences between rd- curves,”ITU SG16 Doc. VCEG-M33, 2001. JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 13

    work page 2001

  64. [64]

    Compact 3d scene representation via self-organizing gaussian grids,

    W. Morgenstern, F. Barthel, A. Hilsmann, and P. Eisert, “Compact 3d scene representation via self-organizing gaussian grids,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 18–34

    work page 2024

  65. [65]

    Compgs: Efficient 3d scene representation via compressed gaussian splatting,

    X. Liu, X. Wu, P. Zhang, S. Wang, Z. Li, and S. Kwong, “Compgs: Efficient 3d scene representation via compressed gaussian splatting,” in Proceedings of the 32nd ACM International Conference on Multimedia, 2024, pp. 2936–2944

    work page 2024

  66. [66]

    Fast feedforward 3d gaussian splatting compression,

    Y . Chen, Q. Wu, M. Li, W. Lin, M. Harandi, and J. Cai, “Fast feedforward 3d gaussian splatting compression,” inThe Thirteenth International Conference on Learning Representations, 2025. [Online]. Available: https://openreview.net/forum?id=DCandSZ2F1 JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2021 14 –Supplementary Material– TABLE V: Results of ...

    work page 2025