pith. machine review for the scientific record. sign in

arxiv: 2605.02726 · v1 · submitted 2026-05-04 · 📡 eess.IV

Recognition: unknown

Cool-chic 5.0: Faster Encoding and Inter-Feature Entropy Modeling for Overfitted Image Compression

Th\'eo Ladune , Pierrick Philippe , Pierre Jaffuer , Th\'eophile Blard , Sylvain Kervadec , F\'elix Henry , Gordon Clare
Authors on Pith no claims yet

Pith reviewed 2026-05-08 02:14 UTC · model grok-4.3

classification 📡 eess.IV
keywords overfitted image compressionCool-chicinter-feature entropy modelingdecoder architecturerate-distortion performanceH.266/VVCautoencoderslow-complexity decoding
0
0 comments X

The pith

Cool-chic 5.0 updates the decoder architecture and optimization to deliver better image compression with one-tenth the encoding iterations of prior overfitted methods.

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

The paper advances overfitted image compression, where a decoder is learned specifically for each image at encode time. It claims that adding inter-feature entropy modeling to the decoder and refining the optimization process produces higher compression performance than earlier versions of the same family while cutting encoding steps by a factor of ten. A reader would care because the result keeps decoding extremely lightweight compared with general-purpose learned codecs yet still reaches or exceeds the rates of both conventional standards and modern autoencoders.

Core claim

Cool-chic 5.0 reaches superior rate-distortion performance by combining an updated decoder that incorporates inter-feature entropy modeling with a more efficient optimization schedule, allowing it to surpass all previous overfitted codecs using only one-tenth as many encoding iterations, to reduce bitrate by 11 percent relative to H.266/VVC, and to remain competitive with autoencoders such as MLIC++ at 250 times lower decoding complexity.

What carries the argument

Inter-feature entropy modeling inside the content-tailored decoder, which captures statistical dependencies across feature maps to produce tighter probability estimates for arithmetic coding.

If this is right

  • Only one-tenth the encoding iterations are needed to exceed the compression of earlier overfitted codecs.
  • Bitrate drops by 11 percent compared with the conventional H.266/VVC standard under the reported test conditions.
  • Rate-distortion performance stays competitive with modern autoencoders while decoding complexity falls by a factor of 250.
  • The full implementation is released as open source, enabling direct replication and extension.

Where Pith is reading between the lines

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

  • The reduced encoding cost could make content-specific compression practical on devices where compute time is limited but high quality is required.
  • The same inter-feature modeling pattern might be applied to other media such as video to obtain similar complexity reductions.
  • Public code release allows independent verification and possible integration into existing compression pipelines.

Load-bearing premise

The measured gains come from the stated decoder changes and optimization schedule rather than from any unmentioned differences in training data, test conditions, or implementation details.

What would settle it

Repeating the experiments while restricting Cool-chic 5.0 to the same number of encoding iterations used by the previous best overfitted codec and finding that the rate advantage disappears would falsify the central claim.

Figures

Figures reproduced from arXiv: 2605.02726 by F\'elix Henry, Gordon Clare, Pierre Jaffuer, Pierrick Philippe, Sylvain Kervadec, Th\'eo Ladune, Th\'eophile Blard.

Figure 1
Figure 1. Figure 1: Compression performance and encoding complexity of view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the proposed decoder for Cool-chic 5.0, with view at source ↗
Figure 3
Figure 3. Figure 3: Autoregressive probability model (ARM) and Inter view at source ↗
Figure 4
Figure 4. Figure 4: Architecture of the ARM fψ and Synthesis fθ neural networks. Both networks feature the linear residual stabilizer view at source ↗
Figure 5
Figure 5. Figure 5: Architecture of the upsampling fυ. Conv and TConv denote convolution and transposed convolution layers. Then, the synthesis transform fθ maps the dense represen￾tation to the decoded image: xˆ = fθ(zˆ), where xˆ shape is (C, H, W), (8) where C is the number of image channels, typically 3 for RGB. The synthesis architecture is presented in Fig. 4b. The trunk branch is composed of point-wise (1 × 1 kernel) c… view at source ↗
Figure 6
Figure 6. Figure 6: BD-rate versus VVC against the decoding complexity in multiplication accumulations (MAC) per pixel. view at source ↗
Figure 7
Figure 7. Figure 7: Sequence-wise BD-rate of the proposed system versus VVC (VTM 28.3) on the CLIC20 professional validation dataset. view at source ↗
Figure 8
Figure 8. Figure 8: Rate-distortion graph on 3 images from the CLIC20 professional validation dataset. view at source ↗
Figure 9
Figure 9. Figure 9: Size of the compressed neural network parameters view at source ↗
read the original abstract

Overfitted codecs compress an image by learning a decoder tailored to the content during the encoding. As such, they trade increased encoding complexity for strong compression performance and low decoding complexity. This work introduces Cool-chic 5.0, the latest version in the Cool-chic series of overfitted codecs, featuring an updated decoder architecture and an improved optimization process. Cool-chic 5.0 outperforms all overfitted codecs with 10 times less encoding iterations. It offers -11% rate reduction compared to the state-of-the-art conventional codec H.266/VVC. It is also competitive with modern autoencoders such as MLIC++ while featuring a decoding complexity 250 times lower. This work is made open-source at https://github.com/Orange-OpenSource/Cool-Chic.

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 / 3 minor

Summary. The manuscript presents Cool-chic 5.0, an updated overfitted image codec that incorporates a revised decoder architecture, inter-feature entropy modeling, and refined optimization procedures. It claims to surpass prior overfitted codecs in rate-distortion performance while requiring only one-tenth the encoding iterations, to deliver an 11% rate saving relative to H.266/VVC, and to remain competitive with learned autoencoders such as MLIC++ at 250 times lower decoding complexity. The implementation is released as open-source software.

Significance. If the reported gains are reproducible under matched experimental conditions, the work meaningfully improves the practicality of overfitted compression by simultaneously lowering encoding cost and raising compression efficiency. The open-source release and the explicit architectural diagrams plus loss formulations constitute a clear strength, enabling direct verification and extension by the community.

major comments (2)
  1. [Section 4 (Experimental results) and Section 3.2 (Inter-feature entropy modeling)] The central performance claims rest on the inter-feature entropy model and the updated decoder; however, the manuscript does not provide an ablation that isolates the contribution of each change (e.g., entropy modeling alone versus decoder redesign) to the observed 10× iteration reduction and rate savings. Without this breakdown, it is difficult to attribute the gains precisely to the stated innovations.
  2. [Table 2 and Figure 5] Table 2 and the associated RD curves compare against H.266/VVC and MLIC++; the text should explicitly state the VVC encoder configuration (preset, intra-only flag, rate-control mode) and confirm that all methods were evaluated on identical test images and with the same distortion metric (PSNR or MS-SSIM) at the same operating points.
minor comments (3)
  1. [Abstract and Section 4.1] The abstract states “10 times less encoding iterations” without referencing the exact baseline iteration count used for the comparison; this detail should appear in the main text near the first mention of the result.
  2. [Section 3.2 and Equation (5)] Notation for the entropy parameters (e.g., the definition of the inter-feature context vector) is introduced in Section 3.2 but is not consistently reused in the loss-function equation; a single consolidated notation table would improve readability.
  3. [Section 5 (Conclusion)] The open-source repository link is given, but the manuscript should include a brief statement confirming that the released code reproduces the exact numbers reported in Tables 1–3.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive assessment and recommendation for minor revision. The comments highlight important aspects for clarity and attribution, which we address below. We will incorporate the requested details and additions in the revised manuscript.

read point-by-point responses

  1. Referee: [Section 4 (Experimental results) and Section 3.2 (Inter-feature entropy modeling)] The central performance claims rest on the inter-feature entropy model and the updated decoder; however, the manuscript does not provide an ablation that isolates the contribution of each change (e.g., entropy modeling alone versus decoder redesign) to the observed 10× iteration reduction and rate savings. Without this breakdown, it is difficult to attribute the gains precisely to the stated innovations.

    Authors: We agree that an explicit ablation isolating the inter-feature entropy modeling from the decoder redesign would improve attribution of the gains in rate savings and encoding iterations. The current manuscript presents the combined Cool-chic 5.0 system, but we will add a dedicated ablation subsection (or table) in the revised version. This will report separate results for the entropy model alone, the decoder changes alone, and their combination, using the same test conditions to quantify each component's contribution to the 10× iteration reduction and BD-rate improvements. revision: yes

  2. Referee: [Table 2 and Figure 5] Table 2 and the associated RD curves compare against H.266/VVC and MLIC++; the text should explicitly state the VVC encoder configuration (preset, intra-only flag, rate-control mode) and confirm that all methods were evaluated on identical test images and with the same distortion metric (PSNR or MS-SSIM) at the same operating points.

    Authors: We appreciate this request for explicit experimental details. In the revised manuscript, we will add a clear statement in Section 4 (and in the captions of Table 2 and Figure 5) specifying the VVC configuration: the 'faster' preset, intra-only mode enabled, and constant quality rate control with QP values matched to the target rates. We confirm that all codecs (including Cool-chic 5.0, prior overfitted methods, and MLIC++) were evaluated on identical test images from the standard datasets using PSNR as the distortion metric at the same operating points. These clarifications will be inserted to ensure full reproducibility. revision: yes

Circularity Check

0 steps flagged

Empirical engineering paper; no circular derivation

full rationale

The paper introduces Cool-chic 5.0 as an updated overfitted image codec with new decoder architecture, inter-feature entropy modeling, and optimization tweaks. All central claims (10x fewer iterations, -11% rate vs H.266/VVC, competitiveness with MLIC++) are supported by reported RD tables, complexity measurements, and an open-source repository for direct reproduction. No mathematical derivation chain exists that could reduce predictions to fitted inputs or self-citations by construction. Minor series self-reference is present but not load-bearing for any result.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no specific free parameters, axioms, or invented entities can be identified from the provided information.

pith-pipeline@v0.9.0 · 5463 in / 1101 out tokens · 37425 ms · 2026-05-08T02:14:10.275268+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

47 extracted references · 20 canonical work pages · 1 internal anchor

  1. [1]

    Cool- chic: Coordinate-based low complexity hierarchical image codec,

    T. Ladune, P. Philippe, F. Henry, G. Clare, and T. Leguay, “Cool- chic: Coordinate-based low complexity hierarchical image codec,” in Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), October 2023, pp. 13 515–13 522

    2023

  2. [2]

    Mlic++: Linear complexity multi-reference entropy modeling for learned image compression,

    W. Jiang, J. Yang, Y . Zhai, F. Gao, and R. Wang, “Mlic++: Linear complexity multi-reference entropy modeling for learned image compression,”ACM Trans. Multimedia Comput. Commun. Appl., vol. 21, no. 5, May 2025. [Online]. Available: https://doi.org/10.1145/3719011

    work page doi:10.1145/3719011 2025

  3. [3]

    Learned image compression with hierarchical progressive context modeling,

    Y . Li, H. Zhang, L. Li, and D. Liu, “Learned image compression with hierarchical progressive context modeling,”CoRR, vol. abs/2507.19125,

    work page arXiv

  4. [4]

    Learned image compression with hierarchical progressive context modeling,

    [Online]. Available: https://doi.org/10.48550/arXiv.2507.19125

    work page doi:10.48550/arxiv.2507.19125

  5. [5]

    Improved encoding for overfitted video codecs,

    P. Philippe, T. Ladune, G. Clare, F. Henry, T. Blard, and T. Leguay, “Upsampling improvement for overfitted neural coding,” inIEEE International Symposium on Circuits and Systems, ISCAS 2025, London, United Kingdom, May 25-28, 2025. IEEE, 2025, pp. 1–5. [Online]. Available: https://doi.org/10.1109/ISCAS56072.2025.11044014

    work page doi:10.1109/iscas56072.2025.11044014 2025

  6. [6]

    MoRIC: A modular region-based implicit codec for image compression,

    G. Li, H. Wu, and D. Gunduz, “MoRIC: A modular region-based implicit codec for image compression,” inThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025. [Online]. Available: https://openreview.net/forum?id=IFjjzfkC65

    2025

  7. [7]

    Lotterycodec: Searching the implicit representation in a random network for low-complexity image compression,

    H. Wu, G. Chen, P. L. Dragotti, and D. G ¨und¨uz, “Lotterycodec: Searching the implicit representation in a random network for low-complexity image compression,” 2025. [Online]. Available: https: //arxiv.org/abs/2507.01204

    work page arXiv 2025

  8. [8]

    The Cool-chic image and video codec,

    O. Research, “The Cool-chic image and video codec,” https://github.com/Orange-OpenSource/Cool-Chic

  9. [9]

    C3: High-performance and low-complexity neural compression from a single image or video,

    H. Kim, M. Bauer, L. Theis, J. R. Schwarz, and E. Dupont, “C3: High-performance and low-complexity neural compression from a single image or video,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2024, pp. 9347– 9358

    2024

  10. [10]

    CLIC Challenge on Learned Image Coding 2020,

    CLIC20, “CLIC Challenge on Learned Image Coding 2020,” http://clic.compression.cc/2021/tasks/index.html, 2020

    2020

  11. [11]

    Implicit neural representations with periodic activation functions,

    V . Sitzmann, J. N. P. Martel, A. W. Bergman, D. B. Lindell, and G. Wetzstein, “Implicit neural representations with periodic activation functions,” inAdvances in Neural Information Processing Systems 33: Annual Conference on Neural Information Processing Systems 2020, NeurIPS 2020, December 6-12, 2020, virtual, H. Larochelle, M. Ranzato, R. Hadsell, M. B...

    2020

  12. [12]

    Coin: Compression with implicit neural representations.arXiv preprint arXiv:2103.03123, 2021

    E. Dupont, A. Golinski, M. Alizadeh, Y . W. Teh, and A. Doucet, “Coin: Compression with implicit neural representations,”ArXiv, vol. abs/2103.03123, 2021. [Online]. Available: https://api.semanticscholar. org/CorpusID:232110691

    work page arXiv 2021

  13. [13]

    COIN++: neural compression across modalities,

    E. Dupont, H. Loya, M. Alizadeh, A. Golinski, Y . W. Teh, and A. Doucet, “COIN++: neural compression across modalities,” Trans. Mach. Learn. Res., vol. 2022, 2022. [Online]. Available: https://openreview.net/forum?id=NXB0rEM2Tq

    2022

  14. [14]

    Implicit neural representations for image compression,

    Y . Str ¨umpler, J. Postels, R. Yang, L. V . Gool, and F. Tombari, “Implicit neural representations for image compression,” inComputer Vision - ECCV 2022 - 17th European Conference, Tel Aviv, Israel, October 23-27, 2022, Proceedings, Part XXVI, ser. Lecture Notes in Computer Science, S. Avidan, G. J. Brostow, M. Ciss ´e, G. M. Farinella, and T. Hassner, E...

    work page doi:10.1007/978-3-031-19809-0 2022

  15. [15]

    Low-complexity overfitted neural image codec,

    T. Leguay, T. Ladune, P. Philippe, G. Clare, F. Henry, and O. D ´eforges, “Low-complexity overfitted neural image codec,” in2023 IEEE 25th International Workshop on Multimedia Signal Processing (MMSP), 2023, pp. 1–6

    2023

  16. [16]

    Multiresolution contexts for implicit neural codecs,

    H. B. Dogaroglu, C. E. Wiedemann, and E. Steinbach, “Multiresolution contexts for implicit neural codecs,” in2025 Picture Coding Symposium (PCS 2025), Dec 2025

    2025

  17. [17]

    Hypercool: Reducing encoding cost in overfitted codecs with hypernetworks,

    P. B. Tatch ´e, T. Aczel, T. Ladune, and R. Wattenhofer, “Hypercool: Reducing encoding cost in overfitted codecs with hypernetworks,” in AAAI 2026 Workshop on Machine Learning for Wireless Communication and Networks (ML4Wireless), 2026. [Online]. Available: https: //openreview.net/forum?id=nhl96GVc6E

    2026

  18. [18]

    Fitted neural lossless image compres- sion,

    Z. Zhang, Z. Chen, and S. Liu, “Fitted neural lossless image compres- sion,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2025, pp. 23 249–23 258

    2025

  19. [19]

    Progressive COOL- CHIC: efficient decoding for dual-resolution images,

    M. Benjak, Y . Chen, W. Peng, and J. Ostermann, “Progressive COOL- CHIC: efficient decoding for dual-resolution images,” inInternational Conference on Visual Communications and Image Processing, VCIP 2025, Klagenfurt, Austria, December 1-4, 2025. IEEE, 2025, pp. 1–5. [Online]. Available: https://doi.org/10.1109/VCIP67698.2025.11396837

    work page doi:10.1109/vcip67698.2025.11396837 2025

  20. [20]

    Scalable COOL-CHIC: dual-resolution images from a single bitstream,

    ——, “Scalable COOL-CHIC: dual-resolution images from a single bitstream,” inPicture Coding Symposium, PCS 2025, Aachen, Germany, December 8-11, 2025. IEEE, 2025, pp. 1–5. [Online]. Available: https://doi.org/10.1109/PCS65673.2025.11417678

    work page doi:10.1109/pcs65673.2025.11417678 2025

  21. [21]

    Objects disentangled implicit neural representation for image coding,

    C. Lin, Y . Wu, Y . Li, J. Li, K. Zhang, and L. Zhang, “Objects disentangled implicit neural representation for image coding,” in Proceedings of the 3rd International Workshop on Multimedia Content Generation and Evaluation: New Methods and Practice, ser. McGE ’25. New York, NY , USA: Association for Computing Machinery, 2025, p. 147–155. [Online]. Availa...

    work page doi:10.1145/3746278.3759393 2025

  22. [22]

    Good, cheap, and fast: Overfitted image compression with wasserstein distortion,

    J. Ball ´e, L. Versari, E. Dupont, H. Kim, and M. Bauer, “Good, cheap, and fast: Overfitted image compression with wasserstein distortion,” 2025, https://arxiv.org/abs/2412.00505

    work page arXiv 2025

  23. [23]

    Perceptually optimised cool-chic for CLIC 2025,

    P. Philippe, T. Ladune, G. Clare, and F. E. Henry, “Perceptually optimised cool-chic for CLIC 2025,” in7th Challenge on Learned Image Compression, 2025. [Online]. Available: https://openreview.net/ forum?id=7S19tNWnRO

    2025

  24. [24]

    Nerv: neural representations for videos,

    H. Chen, B. He, H. Wang, Y . Ren, S.-N. Lim, and A. Shrivastava, “Nerv: neural representations for videos,” inProceedings of the 35th International Conference on Neural Information Processing Systems, ser. NIPS ’21. Red Hook, NY , USA: Curran Associates Inc., 2024

    2024

  25. [25]

    Hnerv: A hybrid neural representation for videos,

    H. Chen, M. Gwilliam, S.-N. Lim, and A. Shrivastava, “Hnerv: A hybrid neural representation for videos,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023

    2023

  26. [26]

    Hinerv: video compression with hierarchical encoding-based neural representation,

    H. M. Kwan, G. Gao, F. Zhang, A. Gower, and D. Bull, “Hinerv: video compression with hierarchical encoding-based neural representation,” in Proceedings of the 37th International Conference on Neural Information 11 Processing Systems, ser. NIPS ’23. Red Hook, NY , USA: Curran Associates Inc., 2024

    2024

  27. [27]

    Ffnerv: Flow-guided frame- wise neural representations for videos,

    J. C. Lee, D. Rho, J. H. Ko, and E. Park, “Ffnerv: Flow-guided frame- wise neural representations for videos,” inProceedings of the 31st ACM International Conference on Multimedia, ser. MM ’23. New York, NY , USA: Association for Computing Machinery, 2023, p. 7859–7870. [Online]. Available: https://doi.org/10.1145/3581783.3612444

    work page doi:10.1145/3581783.3612444 2023

  28. [28]

    Nvrc: Neural video representation compression,

    H. M. Kwan, G. Gao, F. Zhang, A. Gower, and D. Bull, “Nvrc: Neural video representation compression,” 2024. [Online]. Available: https://arxiv.org/abs/2409.07414

    work page arXiv 2024

  29. [29]

    Cool-chic video: Learned video coding with 800 parameters,

    T. Leguay, T. Ladune, P. Philippe, and O. D ´eforges, “Cool-chic video: Learned video coding with 800 parameters,” inData Compression Con- ference, DCC 2024, Snowbird, UT, USA, March 19-22, 2024, A. Bilgin, J. E. Fowler, J. Serra-Sagrist `a, Y . Ye, and J. A. Storer, Eds. IEEE, 2024, pp. 23–32, https://doi.org/10.1109/DCC58796.2024.00010

    work page doi:10.1109/dcc58796.2024.00010 2024

  30. [30]

    Improved encoding for overfitted video codecs,

    ——, “Improved encoding for overfitted video codecs,” inIEEE International Symposium on Circuits and Systems, ISCAS 2025, London, United Kingdom, May 25-28, 2025. IEEE, 2025, pp. 1–5. [Online]. Available: https://doi.org/10.1109/ISCAS56072.2025.11043596

    work page doi:10.1109/iscas56072.2025.11043596 2025

  31. [31]

    Cnvc: A compact neural video codec with instance-level adaptation,

    Y . Li, C. Lin, J. Li, K. Zhang, and L. Zhang, “Cnvc: A compact neural video codec with instance-level adaptation,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 36, no. 4, pp. 5525– 5537, 2026

    2026

  32. [32]

    SOAP: improving and stabilizing shampoo using adam for language modeling,

    N. Vyas, D. Morwani, R. Zhao, I. Shapira, D. Brandfonbrener, L. Janson, and S. M. Kakade, “SOAP: improving and stabilizing shampoo using adam for language modeling,” inThe Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24-28, 2025. OpenReview.net, 2025. [Online]. Available: https://openreview.net/forum?id=IDxZhXrpNf

    2025

  33. [33]

    Adam: A Method for Stochastic Optimization

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” in3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings, Y . Bengio and Y . LeCun, Eds., 2015. [Online]. Available: http://arxiv.org/abs/1412.6980

    work page internal anchor Pith review arXiv 2015

  34. [34]

    Universally quantized neural compression,

    E. Agustsson and L. Theis, “Universally quantized neural compression,” inAdvances in Neural Information Processing Systems 33: Annual Conference on Neural Information Processing Systems 2020, NeurIPS 2020, December 6-12, 2020, virtual, H. Larochelle, M. Ranzato, R. Hadsell, M. Balcan, and H. Lin, Eds.,

    2020

  35. [35]

    Available: https://proceedings.neurips.cc/paper/2020/ hash/92049debbe566ca5782a3045cf300a3c-Abstract.html

    [Online]. Available: https://proceedings.neurips.cc/paper/2020/ hash/92049debbe566ca5782a3045cf300a3c-Abstract.html

    2020

  36. [36]

    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,” in Advances in Neural Information Processing Systems, S. Bengio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa-Bianchi, and R. Garnett, Eds., vol. 31. Curran Associates, Inc., 2018. [Online]. Available: https://proceedings.neurips.cc/p...

    2018

  37. [37]

    Variational image compression with a scale hyperprior,

    J. Ball ´e, D. Minnen, S. Singh, S. J. Hwang, and N. Johnston, “Variational image compression with a scale hyperprior,” in6th International Conference on Learning Representations, ICLR 2018, Vancouver, BC, Canada, April 30 - May 3, 2018, Conference Track Proceedings. OpenReview.net, 2018. [Online]. Available: https://openreview.net/forum?id=rkcQFMZRb

    2018

  38. [38]

    Deep Residual Learning for Image Recognition , isbn =

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in2016 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016, Las Vegas, NV , USA, June 27-30, 2016. IEEE Computer Society, 2016, pp. 770–778. [Online]. Available: https://doi.org/10.1109/CVPR.2016.90

    work page doi:10.1109/cvpr.2016.90 2016

  39. [39]

    Lossy image compres- sion with compressive autoencoders,

    L. Theis, W. Shi, A. Cunningham, and F. Husz ´ar, “Lossy image compres- sion with compressive autoencoders,”ArXiv, vol. abs/1703.00395, 2017. [Online]. Available: https://api.semanticscholar.org/CorpusID:8394195

    work page arXiv 2017

  40. [40]

    Soft then hard: Rethinking the quantization in neural image compression,

    Z. Guo, Z. Zhang, R. Feng, and Z. Chen, “Soft then hard: Rethinking the quantization in neural image compression,” in International Conference on Machine Learning, 2021. [Online]. Available: https://api.semanticscholar.org/CorpusID:233210102

    2021

  41. [41]

    Leveraging second-order curvature for efficient learned image compression: Theory and empirical evidence,

    Y . Zhang and F. Zhu, “Leveraging second-order curvature for efficient learned image compression: Theory and empirical evidence,”CoRR, vol. abs/2601.20769, 2026. [Online]. Available: https://doi.org/10. 48550/arXiv.2601.20769

    work page arXiv 2026

  42. [42]

    Learned image compression with discretized gaussian mixture likelihoods and attention modules,

    Z. Cheng, H. Sun, M. Takeuchi, and J. Katto, “Learned image compression with discretized gaussian mixture likelihoods and attention modules,”2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 7936–7945, 2020. [Online]. Available: https://api.semanticscholar.org/CorpusID:209862064

    2020

  43. [43]

    Elic: Efficient learned image compression with unevenly grouped space-channel contextual adaptive coding,

    D. He, Z. Yang, W. Peng, R. Ma, H. Qin, and Y . Wang, “Elic: Efficient learned image compression with unevenly grouped space-channel contextual adaptive coding,”2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 5708–5717, 2022. [Online]. Available: https://api.semanticscholar.org/CorpusID:247594672

    2022

  44. [44]

    Towards practical real-time neural video compression,

    Z. Jia, B. Li, J. Li, W. Xie, L. Qi, H. Li, and Y . Lu, “Towards practical real-time neural video compression,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2025, Nashville, TN, USA, June 11-25, 2024, 2025

    2025

  45. [45]

    Kodak image dataset,

    “Kodak image dataset,”http://r0k.us/graphics/kodak/

  46. [46]

    Overview of the Versatile Video Coding (VVC) standard and its applications,

    B. Bross, Y .-K. Wang, Y . Ye, S. Liu, J. Chen, G. J. Sullivan, and J.-R. Ohm, “Overview of the Versatile Video Coding (VVC) standard and its applications,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 31, no. 10, pp. 3736–3764, 2021

    2021

  47. [47]

    Low-rank quantization-aware training for LLMs

    Y . Bondarenko, R. D. Chiaro, and M. Nagel, “Low-rank quantization- aware training for llms,”ArXiv, vol. abs/2406.06385, 2024. [Online]. Available: https://api.semanticscholar.org/CorpusID:270370870

    work page arXiv 2024