pith. sign in

arxiv: 2605.20672 · v1 · pith:QCXW6XDVnew · submitted 2026-05-20 · 📡 eess.IV

LANCE: Locally Adaptive Neural Context Estimation for Overfitted Image Compression

Martin Benjak , J\"orn Ostermann This is my paper

Pith reviewed 2026-05-21 02:33 UTC · model grok-4.3

classification 📡 eess.IV
keywords overfitted image compressionspatial hyperpriorentropy model adaptationneural context estimationpredictive codingMedian Edge Detectorrate-distortion performancelocal image statistics
0
0 comments X

The pith

A forward-signaled spatial hyperprior lets overfitted image compressors adapt their entropy models to local statistics with low 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 improve overfitted image compression by replacing global entropy parameters with regional adaptation. Standard methods struggle when image content changes across the frame, leading to inefficient coding in mixed areas. LANCE adds a spatial hyperprior that is predicted from already decoded pixels using a fixed edge detector plus a small learned model. This keeps extra signaling cheap while letting the entropy coder adjust per region. If the approach holds, it produces modest but consistent rate savings on common test sets without raising decoder cost much.

Core claim

LANCE extends overfitted frameworks such as Cool-Chic by introducing a spatial hyperprior that is signaled forward using predictive coding. The scheme combines a static Median Edge Detector with a lightweight learned context model to generate a map of local statistics. This map then modulates the entropy parameters on a per-region basis. Experiments report BD-rate reductions of 1.40 percent on Kodak and 1.97 percent on CLIC 2020 at the higher complexity end, rising to 2.41 percent and 2.99 percent at the lower end, across decoder complexities of 606 to 1481 MAC per pixel. The hyperprior is shown to group image areas that share similar statistics.

What carries the argument

Spatial hyperprior signaled by predictive coding that combines a Median Edge Detector with a lightweight learned context model to adapt the entropy model regionally.

If this is right

  • Delivers positive BD-rate gains over global-parameter baselines on Kodak and CLIC at both high and low decoder complexity.
  • Enables automatic, content-driven segmentation of regions for entropy modeling.
  • Keeps added decoder operations within a bounded MAC-per-pixel range.
  • Reduces signaling cost through the use of a static edge detector plus minimal learned prediction.

Where Pith is reading between the lines

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

  • The same predictive signaling idea could be tested on video sequences where content varies across both space and time.
  • Pairing the hyperprior with existing learned entropy models might further shrink overhead on high-resolution images.
  • The segmentation behavior suggests the method could be useful in domains that need content-aware bit allocation without manual masks.

Load-bearing premise

The spatial hyperprior can divide the image into regions of similar statistics without the cost of signaling that map exceeding the entropy savings it produces.

What would settle it

Run the method on a new image set and measure whether the bits used to transmit the hyperprior are larger than the reduction in bits needed for the main image data; if overhead dominates, net rate does not improve.

Figures

Figures reproduced from arXiv: 2605.20672 by J\"orn Ostermann, Martin Benjak.

Figure 1
Figure 1. Figure 1: Rate savings in terms of BD-rate versus VVC (VTM 23.14) over [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Decoding process of LANCE: First the spatial hyperprior [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Visualization of the used N-tap context window configurations for the context models Cξ and Cϕ. The blue element is currently being en- or decoded. Gray elements are still unknown to the decoder-side. C. Context Modeling using LANCE To understand why context modeling is necessary, we can look at equation (5) from another view: H(v; q; p) = Ev∼q[− log2 p(v)] = Ev∼q[− log2 p(v) + log2 q(v) − log2 q(v)] = DKL… view at source ↗
Figure 4
Figure 4. Figure 4: Histogram of the BD-rate between 1 million random permutations of [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: BD-rate of our method LANCE versus Cool-Chic 4.0 [8] over the [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Rate-distortion curves evaluated on Kodak [43] (left) and CLIC 2020 professional validation [12] (right) comparing our method LANCE to other [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Rate savings in terms of BD-rate versus VVC (VTM 23.14) over [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Per-image BD-rates of our method LANCE versus Cool-Chic 4.0 [8] evaluated on the Kodak dataset [43]. LANCE is tested with all its modules [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Visualization of normalized spatial hyperprior [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗
read the original abstract

This paper introduces Locally Adaptive Neural Context Estimation (LANCE), a novel extension for overfitted image compression (OIC) frameworks like Cool-Chic. While traditional OIC methods rely on lightweight autoregressive networks with globally signaled parameters, they struggle with non-stationary image statistics. LANCE addresses this by incorporating a forward-signaled spatial hyperprior that enables regional adaptation of the entropy model. To minimize overhead, we employ a predictive coding scheme that combines a static Median Edge Detector (MED) with a lightweight learned context model. Experiments demonstrate that LANCE achieves BD-rate reductions of 1.40% on the Kodak dataset and 1.97% on CLIC 2020 over Cool-Chic 4.0 at the high end of our decoder complexity range of 606-1481 MAC/pixel. At the low end of the complexity range, we outperform Cool-Chic 4.0 by 2.41% and 2.99% on Kodak and CLIC, respectively. Qualitative analysis reveals that the learned spatial hyperprior effectively segments image regions into areas of similar image statistics, providing an automated, content-aware adaptation layer.

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 introduces Locally Adaptive Neural Context Estimation (LANCE) as an extension to overfitted image compression frameworks such as Cool-Chic. It adds a forward-signaled spatial hyperprior that enables regional adaptation of the entropy model to non-stationary image statistics. To control overhead, the hyperprior is signaled via a predictive coder combining a static Median Edge Detector (MED) with a lightweight learned context model. The central empirical claim, supported by experiments on Kodak and CLIC 2020, is that LANCE delivers BD-rate reductions of 1.40% and 1.97% (high-complexity end of the 606–1481 MAC/pixel decoder range) and 2.41% and 2.99% (low-complexity end) relative to Cool-Chic 4.0, accompanied by qualitative evidence that the hyperprior segments regions of similar statistics.

Significance. If the net BD-rate gains survive explicit isolation of hyperprior signaling cost from the main latent stream, the contribution would be a practical advance in locally adaptive entropy modeling for overfitted codecs. The predictive-coding approach that re-uses MED plus a lightweight network is a concrete engineering strength that keeps overhead manageable, and the reported segmentation behavior indicates the hyperprior captures content structure without prohibitive side information. This could inform subsequent designs of content-aware OIC systems.

major comments (2)
  1. [§4.2 and Table 2] §4.2 and Table 2: the reported BD-rate figures compare LANCE against Cool-Chic 4.0 but contain no ablation that disables the spatial hyperprior while holding total decoder complexity fixed at the same MAC/pixel operating points. Without this isolation it is impossible to confirm that entropy savings exceed the added signaling cost of the forward-signaled hyperprior parameters.
  2. [§3.2] §3.2: the predictive-coding description (MED + lightweight context model) does not supply a per-component rate breakdown or an equation that separates hyperprior bits from the main latent stream. Consequently the claim that “overhead remains small enough for positive net BD-rate” rests on aggregate measurements whose sensitivity to hyperprior cost cannot be assessed.
minor comments (2)
  1. [Abstract and §4.1] The complexity range 606–1481 MAC/pixel is stated without an explicit definition of whether the count is decoder-only, includes the hyperprior network, or is measured per pixel versus per image; a short clarifying sentence or footnote would remove ambiguity.
  2. [§5] Figure 4 (qualitative segmentation) would be strengthened by a quantitative metric such as average intra-region variance or edge-alignment score to support the textual claim that the hyperprior “effectively segments image regions.”

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and for recognizing the engineering merits of the predictive coding scheme. We provide point-by-point responses to the major comments below.

read point-by-point responses

  1. Referee: [§4.2 and Table 2] §4.2 and Table 2: the reported BD-rate figures compare LANCE against Cool-Chic 4.0 but contain no ablation that disables the spatial hyperprior while holding total decoder complexity fixed at the same MAC/pixel operating points. Without this isolation it is impossible to confirm that entropy savings exceed the added signaling cost of the forward-signaled hyperprior parameters.

    Authors: The reported operating points maintain fixed total decoder complexity, which for LANCE includes the MAC/pixel cost of the hyperprior decoding network and predictive coder. Since LANCE achieves superior BD-rate performance within these matched complexity ranges, the net entropy savings from regional adaptation exceed the combined rate and complexity overhead of the hyperprior. We will add an explicit ablation disabling the hyperprior (with compensatory adjustments to the base model) at the same complexity points in the revised manuscript to further substantiate this. revision: yes

  2. Referee: [§3.2] §3.2: the predictive-coding description (MED + lightweight context model) does not supply a per-component rate breakdown or an equation that separates hyperprior bits from the main latent stream. Consequently the claim that “overhead remains small enough for positive net BD-rate” rests on aggregate measurements whose sensitivity to hyperprior cost cannot be assessed.

    Authors: Section 3.2 details the predictive coding mechanism designed to minimize hyperprior overhead. The BD-rate results are computed from the aggregate rate, which incorporates both streams. To improve transparency, we will include an equation in the revised manuscript that explicitly separates the total rate into the main latent rate and the hyperprior rate. A per-component breakdown can be added from our existing rate logs to allow assessment of overhead sensitivity. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical BD-rate claims rest on external dataset comparisons

full rationale

The paper introduces LANCE as a practical extension to existing OIC frameworks, employing a spatial hyperprior with MED-based predictive coding to adapt entropy models regionally. All reported results consist of measured BD-rate deltas on Kodak and CLIC 2020 against the Cool-Chic 4.0 baseline across fixed complexity ranges; these are direct experimental outcomes rather than quantities derived from equations that reduce to the paper's own fitted parameters or self-citations. No derivation chain equates a claimed prediction to an input by construction, and the method's novelty is validated through ablation-style complexity sweeps and qualitative segmentation observations instead of tautological re-labeling of fitted quantities.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

Review performed on abstract only; therefore the ledger is necessarily incomplete. The central claim rests on the existence of a learnable spatial hyperprior whose signaling cost is offset by entropy savings.

free parameters (1)
  • spatial hyperprior parameters
    Learned weights of the hyperprior network that are fitted during training and signaled per image or per region.
axioms (1)
  • domain assumption Image statistics are sufficiently non-stationary that global context models leave measurable rate-distortion loss.
    Stated in the abstract as the motivation for regional adaptation.
invented entities (1)
  • forward-signaled spatial hyperprior no independent evidence
    purpose: To provide regional adaptation of the entropy model without large overhead.
    New component introduced by the paper; no independent evidence outside the reported experiments is supplied in the abstract.

pith-pipeline@v0.9.0 · 5732 in / 1295 out tokens · 39620 ms · 2026-05-21T02:33:53.816061+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.

Reference graph

Works this paper leans on

45 extracted references · 45 canonical work pages · 1 internal anchor

  1. [1]

    W. B. Pennebaker and J. L. Mitchell,JPEG: Still image data compres- sion standard. Springer Science & Business Media, 1992

    work page 1992

  2. [2]

    Overview of the high efficiency video coding (hevc) standard,

    G. J. Sullivan, J. Ohm, W. Han, and T. Wiegand, “Overview of the high efficiency video coding (hevc) standard,”IEEE Transactions on Circuits and Systems for Video Technology, 2012

    work page 2012

  3. [3]

    Context-based adaptive binary arithmetic coding in the h.264/avc video compression standard,

    D. Marpe, H. Schwarz, and T. Wiegand, “Context-based adaptive binary arithmetic coding in the h.264/avc video compression standard,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 13, no. 7, pp. 620–636, 2003

    work page 2003

  4. [4]

    Coin: Compression with implicit neural representations,

    E. Dupont, A. Goli ´nski, M. Alizadeh, Y . W. Teh, and A. Doucet, “Coin: Compression with implicit neural representations,”arXiv preprint arXiv:2103.03123, 2021

    work page arXiv 2021

  5. [5]

    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, 2023, pp. 13 515–13 522

    work page 2023

  6. [6]

    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,” in2024 Data Compression Conference (DCC). IEEE, 2024, pp. 23–32

    work page 2024

  7. [7]

    Devel- opments in international video coding standardization after avc, with an overview of versatile video coding (vvc),

    B. Bross, J. Chen, J.-R. Ohm, G. J. Sullivan, and Y .-K. Wang, “Devel- opments in international video coding standardization after avc, with an overview of versatile video coding (vvc),”Proceedings of the IEEE, vol. 109, no. 9, pp. 1463–1493, 2021

    work page 2021

  8. [8]

    Improved encoding for overfitted video codecs,

    T. Leguay, T. Ladune, P. Philippe, and O. D ´eforges, “Improved encoding for overfitted video codecs,” in2025 IEEE International Symposium on Circuits and Systems (ISCAS). IEEE, 2025, pp. 1–5

    work page 2025

  9. [9]

    Mlic++: Linear com- plexity multi-reference entropy modeling for learned image compres- sion,

    W. Jiang, J. Yang, Y . Zhai, F. Gao, and R. Wang, “Mlic++: Linear com- plexity multi-reference entropy modeling for learned image compres- sion,”ACM Transactions on Multimedia Computing, Communications and Applications, vol. 21, no. 5, pp. 1–25, 2025

    work page 2025

  10. [10]

    Learned image com- pression with discretized gaussian mixture likelihoods and attention modules,

    Z. Cheng, H. Sun, M. Takeuchi, and J. Katto, “Learned image com- pression with discretized gaussian mixture likelihoods and attention modules,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020

    work page 2020

  11. [11]

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

    G. Li, H. Wu, and D. G ¨und¨uz, “Moric: A modular region-based implicit codec for image compression,”Conference on Neural Information Processing Systems, 2025

    work page 2025

  12. [12]

    Challenge on learned image coding 2020,

    “Challenge on learned image coding 2020,” 2020. [Online]. Available: https://archive.compression.cc/2021/tasks/index.html

    work page 2020

  13. [13]

    End-to-end optimized image compression,

    J. Ball ´e, V . Laparra, and E. P. Simoncelli, “End-to-end optimized image compression,”arXiv preprint, 2016

    work page 2016

  14. [14]

    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,”arXiv preprint, 2018

    work page 2018

  15. [15]

    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, 2018

    work page 2018

  16. [16]

    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, 2024, pp. 9347–9358

    work page 2024

  17. [17]

    Jpeg xl next-generation image compression architecture and coding tools,

    J. Alakuijala, R. Van Asseldonk, S. Boukortt, M. Bruse, I.-M. Coms ,a, M. Firsching, T. Fischbacher, E. Kliuchnikov, S. Gomez, R. Obryket al., “Jpeg xl next-generation image compression architecture and coding tools,” inApplications of digital image processing XLII, vol. 11137. SPIE, 2019, pp. 112–124

    work page 2019

  18. [18]

    Density modeling of images using a generalized normalization transformation,

    J. Ball ´e, V . Laparra, and E. P. Simoncelli, “Density modeling of images using a generalized normalization transformation,”arXiv preprint arXiv:1511.06281, 2015

    work page arXiv 2015

  19. [19]

    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

    work page 2020

  20. [20]

    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,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 5718– 5727

    work page 2022

  21. [21]

    Entroformer: A transformer-based entropy model for learned image compression,

    Y . Qian, M. Lin, X. Sun, Z. Tan, and R. Jin, “Entroformer: A transformer-based entropy model for learned image compression,”arXiv preprint arXiv:2202.05492, 2022

    work page arXiv 2022

  22. [22]

    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,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2025, pp. 18 834–18 843

    work page 2025

  23. [23]

    Joint global and local hierarchical priors for learned image compression,

    J.-H. Kim, B. Heo, and J.-S. Lee, “Joint global and local hierarchical priors for learned image compression,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022, pp. 5992–6001

    work page 2022

  24. [24]

    Diversify, contextualize, and adapt: Efficient entropy modeling for neural image codec,

    J.-H. Kim, S. Kim, W.-H. Lee, and D. Oh, “Diversify, contextualize, and adapt: Efficient entropy modeling for neural image codec,”Advances in Neural Information Processing Systems, vol. 37, pp. 45 956–45 974, 2024

    work page 2024

  25. [25]

    Compositional pattern producing networks: A novel abstraction of development,

    K. O. Stanley, “Compositional pattern producing networks: A novel abstraction of development,”Genetic programming and evolvable ma- chines, vol. 8, no. 2, pp. 131–162, 2007

    work page 2007

  26. [26]

    Deepsdf: Learning continuous signed distance functions for shape rep- resentation,

    J. J. Park, P. Florence, J. Straub, R. Newcombe, and S. Lovegrove, “Deepsdf: Learning continuous signed distance functions for shape rep- resentation,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 165–174

    work page 2019

  27. [27]

    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

  28. [28]

    Implicit neural representations with periodic activation functions,

    V . Sitzmann, J. Martel, A. Bergman, D. Lindell, and G. Wetzstein, “Implicit neural representations with periodic activation functions,” Advances in neural information processing systems, vol. 33, pp. 7462– 7473, 2020

    work page 2020

  29. [29]

    COIN++: Neural compression across modalities,

    E. Dupont, H. Loya, M. Alizadeh, A. Goli ´nski, Y . W. Teh, and A. Doucet, “Coin++: Neural compression across modalities,”arXiv preprint arXiv:2201.12904, 2022

    work page arXiv 2022

  30. [30]

    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,” inEuropean Conference on Computer Vision. Springer, 2022, pp. 74–91. 13

    work page 2022

  31. [31]

    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). IEEE, 2023, pp. 1–6

    work page 2023

  32. [32]

    Overfitted image coding at reduced complexity,

    T. Blard, T. Ladune, P. Philippe, G. Clare, X. Jiang, and O. D ´eforges, “Overfitted image coding at reduced complexity,” in2024 32nd Eu- ropean Signal Processing Conference (EUSIPCO). IEEE, 2024, pp. 927–931

    work page 2024

  33. [33]

    Upsampling improvement for overfitted neural coding,

    P. Philippe, T. Ladune, G. Clare, F. Henry, T. Blard, and T. Leguay, “Upsampling improvement for overfitted neural coding,”arXiv preprint arXiv:2411.19249, 2024

    work page arXiv 2024

  34. [34]

    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,” in Proceedings of the Computer Vision and Pattern Recognition Confer- ence, 2025, pp. 23 259–23 268

    work page 2025

  35. [35]

    Efficient sub-pixel motion compensation in learned video codecs,

    T. Ladune, T. Leguay, P. Philippe, G. Clare, and F. Henry, “Efficient sub-pixel motion compensation in learned video codecs,”arXiv preprint arXiv:2507.21926, 2025

    work page arXiv 2025

  36. [36]

    Fitted neural lossless image compres- sion,

    Z. Zhang, Z. Chen, and S. Liu, “Fitted neural lossless image compres- sion,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 23 249–23 258

    work page 2025

  37. [37]

    Scalable cool- chic: Dual-resolution images from a single bitstream,

    M. Benjak, Y .-H. Chen, W.-H. Peng, and J. Ostermann, “Scalable cool- chic: Dual-resolution images from a single bitstream,” in2025 Picture Coding Symposium (PCS), 2025

    work page 2025

  38. [38]

    Progressive cool-chic: Efficient decoding for dual-resolution images,

    ——, “Progressive cool-chic: Efficient decoding for dual-resolution images,” in2025 IEEE International Conference on Visual Communi- cations and Image Processing (VCIP), 2025

    work page 2025

  39. [39]

    Universally quantized neural compression,

    E. Agustsson and L. Theis, “Universally quantized neural compression,” Advances in neural information processing systems, vol. 33, pp. 12 367– 12 376, 2020

    work page 2020

  40. [40]

    Lossy Image Compression with Compressive Autoencoders

    L. Theis, W. Shi, A. Cunningham, and F. Husz ´ar, “Lossy image compres- sion with compressive autoencoders,”arXiv preprint arXiv:1703.00395, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  41. [41]

    Loco-i: A low com- plexity, context-based, lossless image compression algorithm,

    M. J. Weinberger, G. Seroussi, and G. Sapiro, “Loco-i: A low com- plexity, context-based, lossless image compression algorithm,” inPro- ceedings of Data Compression Conference-DCC’96. IEEE, 1996, pp. 140–149

    work page 1996

  42. [42]

    The loco-i lossless image compression algorithm: principles and standardization into jpeg-ls,

    M. Weinberger, G. Seroussi, and G. Sapiro, “The loco-i lossless image compression algorithm: principles and standardization into jpeg-ls,” IEEE Transactions on Image Processing, vol. 9, no. 8, pp. 1309–1324, 2000

    work page 2000

  43. [43]

    Kodak image dataset

    “Kodak image dataset.” [Online]. Available: http://r0k.us/graphics/ kodak/

    work page

  44. [44]

    Calculation of average psnr differences between rd- curves,

    G. Bjontegaard, “Calculation of average psnr differences between rd- curves,”ITU-T SG16 VCEG-M33, 2001

    work page 2001

  45. [45]

    Vca: video complexity analyzer,

    V . V . Menon, C. Feldmann, H. Amirpour, M. Ghanbari, and C. Tim- merer, “Vca: video complexity analyzer,” inProceedings of the 13th ACM multimedia systems conference, 2022, pp. 259–264. Martin Benjakreceived the B.Sc. degree in elec- trical engineering from the University of Applied Sciences of Osnabr ¨uck, in 2017, and the M.Sc. degree in electrical eng...

    work page 2022