pith. sign in

arxiv: 2606.01608 · v1 · pith:CJLIMHLQnew · submitted 2026-06-01 · 💻 cs.CV

Exploiting Semantic and Pixel Representations for Ultra-Low Bitrate Image Compression

Hao Wei , Yanhui Zhou , Chenyang Ge , Saeed Anwar , Ajmal Mian This is my paper

Pith reviewed 2026-06-28 15:49 UTC · model grok-4.3

classification 💻 cs.CV
keywords image compressionultra-low bitratediffusion modelssemantic representationspixel fidelityrate-distortion-perceptiontriple-encoder architecture
0
0 comments X

The pith

A triple-encoder diffusion model compensates for a frozen VAE by fusing semantic and distortion features, improving both pixel fidelity and perceptual quality below 0.03 bpp.

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

The paper presents SPRDiff as a way to compress images at bitrates so low that standard methods lose either visual realism or exact pixel match to the source. It builds a triple-encoder stack that adds pretrained semantic and distortion encoders to a frozen VAE, then feeds the richer latents into an entropy model. A separate distortion-aware module produces a coarse structure map plus dual semantic-pixel signals that condition a diffusion decoder. The central aim is to reach a better rate-distortion-perception operating point than prior extreme-compression techniques. Experiments on standard benchmarks show the resulting images stay closer to the originals in both pixel error and human-perceived quality.

Core claim

SPRDiff shows that high-fidelity features from pretrained distortion-oriented and semantic-oriented encoders can compensate for the limited representations extracted by a frozen VAE encoder, thereby improving latent compression and entropy modeling, while a distortion-aware reconstruction module supplies accurate conditional signals that let the diffusion model preserve both main structures and fine pixel-level detail at bitrates below 0.03 bpp.

What carries the argument

Triple-encoder architecture that fuses features from a frozen VAE encoder with pretrained distortion-oriented and semantic-oriented encoders, paired with a distortion-aware reconstruction module that extracts dual semantic-pixel conditions to guide diffusion reconstruction.

If this is right

  • Images reconstructed below 0.03 bpp retain measurable pixel accuracy while also appearing more realistic than reconstructions from earlier ultra-low-bitrate codecs.
  • Latent entropy coding benefits directly from the added semantic and distortion features, lowering the rate needed for a given reconstruction quality.
  • The dual-signal conditioning from the distortion-aware module allows the diffusion stage to correct structural errors without introducing new semantic drift.
  • The overall rate-distortion-perception surface shifts outward compared with methods that optimize only one or two of the three axes.

Where Pith is reading between the lines

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

  • The same compensation pattern could be tested on other frozen generative backbones, such as those used in video or 3D compression, by swapping in domain-matched semantic encoders.
  • If the pretrained encoders are themselves updated on the target domain, the performance gap at ultra-low rates might widen further, offering a route to task-specific extreme compression.
  • The explicit separation of coarse structure from fine conditional signals suggests a natural way to add temporal consistency constraints when extending the method to video sequences.

Load-bearing premise

Features from the two pretrained encoders can supply the information missing from the frozen VAE encoder and thereby improve the latent representation used for compression.

What would settle it

Side-by-side evaluation on benchmark images at bitrates strictly below 0.03 bpp measuring whether SPRDiff simultaneously improves both pixel-wise metrics such as PSNR and perceptual metrics such as LPIPS relative to prior methods.

Figures

Figures reproduced from arXiv: 2606.01608 by Ajmal Mian, Chenyang Ge, Hao Wei, Saeed Anwar, Yanhui Zhou.

Figure 1
Figure 1. Figure 1: Comparison with state-of-the-art compression methods. The value in parentheses indicates the bpp used by the compressed image. The best value is [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of diffusion-based extreme compression methods on the [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The overall framework of the proposed SPRDiff consists of a triple-encoder, a feature fusion block, a latent compression module, and a one [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Visual comparisons of reconstructed images at around 0.02 bpp using [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Reconstruction results at approximately 0.03 bpp. Panel (b) preserves [PITH_FULL_IMAGE:figures/full_fig_p004_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Comparison of extreme image compression methods on benchmark datasets. Lower LPIPS, DISTS, FID, and KID values, and higher PSNR and [PITH_FULL_IMAGE:figures/full_fig_p006_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Visual comparisons with state-of-the-art methods on the CLIC2020 dataset. The value in parentheses denotes the bpp used for compression, with the [PITH_FULL_IMAGE:figures/full_fig_p007_7.png] view at source ↗
Figure 9
Figure 9. Figure 9: Ablation studies on the semantic and pixel-level representation [PITH_FULL_IMAGE:figures/full_fig_p008_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Effectiveness of semantic representation [PITH_FULL_IMAGE:figures/full_fig_p008_10.png] view at source ↗
Figure 8
Figure 8. Figure 8: The rate–distortion–perception validation loss curve for the first [PITH_FULL_IMAGE:figures/full_fig_p008_8.png] view at source ↗
Figure 11
Figure 11. Figure 11: Impact of pixel representation Fp. (b) and (c) are reconstructions using the method without and with pixel guidance at around 0.02 bpp, respectively. Original image (a) Original patch (24) (b) DiffEIC (0.0145) (c) ResULIC (0.0125) (d) RDEIC (0.0166) (e) StableCodec (0.0189) (f) SPRDiff (0.0116) [PITH_FULL_IMAGE:figures/full_fig_p009_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Limitation: Facial reconstruction in complex backgrounds. The value [PITH_FULL_IMAGE:figures/full_fig_p009_12.png] view at source ↗
read the original abstract

Most existing extreme compression methods fail to achieve an optimal rate-distortion-perception trade-off, as they typically prioritize perceptual fidelity and visual realism over pixel-level accuracy. Consequently, the resulting reconstructions often deviate noticeably from the originals. Ultra-low bitrate image compression is therefore crucial-not only for producing extremely compact representations but also for ensuring that reconstructed images remain semantically coherent and faithful to the source at the pixel level. To this end, we propose SPRDiff, a diffusion-based compression method that fully leverages both semantic and pixel representations, thereby enhancing reconstruction fidelity under ultra-low bitrate constraints. Specifically, we develop a triple-encoder architecture that utilizes high-fidelity features from the pretrained distortion-oriented and semantic-oriented encoders to compensate for the limited representations extracted by the frozen VAE encoder, thereby improving latent compression and entropy modeling. To further enhance the reconstruction fidelity of diffusion models, we introduce a distortion-aware reconstruction module with dual feature extraction. This module not only generates a coarse reconstruction that preserves the main structures, but also provides practical and accurate semantic- and pixel-level conditional signals to guide the diffusion model. Extensive experiments on benchmark datasets demonstrate that our method outperforms state-of-the-art approaches in the rate-distortion-perception tradeoff at extremely low bitrates (below 0.03 bpp), effectively preserving both perceptual quality and pixel-wise fidelity in the reconstructed images. We will release the source code and trained models at https://github.com/cshw2021/SPRDiff.

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 / 1 minor

Summary. The manuscript proposes SPRDiff, a diffusion-based image compression framework that employs a triple-encoder architecture (frozen VAE encoder augmented by pretrained distortion-oriented and semantic-oriented encoders) together with a distortion-aware reconstruction module to improve latent compression, entropy modeling, and diffusion-based decoding. The central claim is that this design yields superior rate-distortion-perception trade-offs at bitrates below 0.03 bpp while preserving both perceptual quality and pixel-wise fidelity, with extensive experiments on benchmark datasets demonstrating outperformance over state-of-the-art methods.

Significance. If the experimental support for the triple-encoder compensation mechanism and the reported RD-P gains holds, the work would address a practically relevant regime of extreme compression where both semantic coherence and pixel accuracy matter, potentially informing future diffusion-based codecs.

major comments (2)
  1. [Abstract / triple-encoder description] Abstract / method description of the triple-encoder architecture: the assertion that high-fidelity features from the pretrained distortion-oriented and semantic-oriented encoders compensate for the limited representations of the frozen VAE encoder (thereby improving latent compression and entropy modeling) is presented as load-bearing for the claimed gains, yet the provided material contains no ablation study, quantitative contribution analysis, or comparison isolating this compensation effect versus a standard conditional diffusion baseline.
  2. [Abstract] Abstract: the claim that 'extensive experiments on benchmark datasets demonstrate that our method outperforms state-of-the-art approaches in the rate-distortion-perception tradeoff at extremely low bitrates (below 0.03 bpp)' is made without any reported quantitative results, error bars, dataset statistics, specific baselines, or RD-P curves in the supplied text, rendering the central empirical claim unevaluable.
minor comments (1)
  1. The manuscript states that source code and trained models will be released; this is a positive step for reproducibility and should be confirmed in the camera-ready version.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major point below and commit to revisions that strengthen the manuscript without altering its core claims.

read point-by-point responses

  1. Referee: [Abstract / triple-encoder description] Abstract / method description of the triple-encoder architecture: the assertion that high-fidelity features from the pretrained distortion-oriented and semantic-oriented encoders compensate for the limited representations of the frozen VAE encoder (thereby improving latent compression and entropy modeling) is presented as load-bearing for the claimed gains, yet the provided material contains no ablation study, quantitative contribution analysis, or comparison isolating this compensation effect versus a standard conditional diffusion baseline.

    Authors: We acknowledge that the manuscript does not contain an ablation that isolates the specific compensation effect of the triple-encoder architecture against a standard conditional diffusion baseline. While the overall rate-distortion-perception results are reported in the experiments section, an explicit quantitative breakdown of each encoder's contribution would make the design rationale more transparent. We will add this ablation study, including comparisons with and without the distortion-oriented and semantic-oriented encoders, in the revised manuscript. revision: yes

  2. Referee: [Abstract] Abstract: the claim that 'extensive experiments on benchmark datasets demonstrate that our method outperforms state-of-the-art approaches in the rate-distortion-perception tradeoff at extremely low bitrates (below 0.03 bpp)' is made without any reported quantitative results, error bars, dataset statistics, specific baselines, or RD-P curves in the supplied text, rendering the central empirical claim unevaluable.

    Authors: The abstract summarizes the experimental findings without embedding specific numerical values, which is common but can limit immediate evaluability. The full manuscript contains the requested quantitative results, including RD-P curves, dataset details, and comparisons against listed baselines in Section 4. To address the concern directly, we will revise the abstract to include key quantitative highlights (e.g., specific PSNR, LPIPS, and bitrate values) while remaining within length constraints. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical claims rest on experiments, not self-referential derivations

full rationale

The paper's core claims concern empirical outperformance of SPRDiff on rate-distortion-perception metrics below 0.03 bpp, achieved via a proposed triple-encoder design and distortion-aware module. No equations, fitted parameters, or derivations are shown that reduce these results to inputs by construction (e.g., no self-definitional compensation quantified as a prediction, no self-citation load-bearing uniqueness theorems, and no renaming of known patterns). The compensation assumption is a stated design rationale whose validity is asserted via benchmark results rather than circular logic. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Review is based on the abstract alone. The approach relies on standard properties of diffusion models and pretrained vision encoders from prior literature without introducing new free parameters or invented entities in the provided text.

axioms (1)
  • domain assumption Diffusion models can be effectively conditioned on semantic- and pixel-level signals extracted from a coarse reconstruction to improve fidelity under ultra-low bitrate constraints
    Invoked when describing how the distortion-aware reconstruction module supplies conditional signals to the diffusion model.

pith-pipeline@v0.9.1-grok · 5800 in / 1343 out tokens · 35955 ms · 2026-06-28T15:49:23.156627+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

69 extracted references · 13 canonical work pages · 1 internal anchor

  1. [1]

    The jpeg still picture compression standard,

    G. K. Wallace, “The jpeg still picture compression standard,”Commu- nications of the ACM, vol. 34, no. 4, pp. 30–44, 1991

    1991

  2. [2]

    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

  3. [3]

    Learned image compression with mixed transformer-cnn architectures,

    J. Liu, H. Sun, and J. Katto, “Learned image compression with mixed transformer-cnn architectures,” inProceedings of the IEEE/CVF con- ference on computer vision and pattern recognition, 2023, pp. 14 388– 14 397

    2023

  4. [4]

    Frequency- aware transformer for learned image compression,

    H. Li, S. Li, W. Dai, C. Li, J. Zou, and H. Xiong, “Frequency- aware transformer for learned image compression,”arXiv preprint arXiv:2310.16387, 2023

    work page arXiv 2023

  5. [5]

    Improving statistical fidelity for neural image compression with im- plicit local likelihood models,

    M. J. Muckley, A. El-Nouby, K. Ullrich, H. J ´egou, and J. Verbeek, “Improving statistical fidelity for neural image compression with im- plicit local likelihood models,” inInternational Conference on Machine Learning. PMLR, 2023, pp. 25 426–25 443

    2023

  6. [6]

    Generative latent coding for ultra-low bitrate image compression,

    Z. Jia, J. Li, B. Li, H. Li, and Y . Lu, “Generative latent coding for ultra-low bitrate image compression,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 26 088–26 098. 10

    2024

  7. [7]

    A lightweight model for perceptual image compression via implicit priors,

    H. Wei, Y . Zhou, Y . Jia, C. Ge, S. Anwar, and A. Mian, “A lightweight model for perceptual image compression via implicit priors,”Neural Networks, p. 108279, 2025. [Online]. Available: https://www.sciencedirect.com/science/article/pii/S0893608025011608

    2025

  8. [8]

    Towards extreme image compression with latent feature guidance and diffusion prior,

    Z. Li, Y . Zhou, H. Wei, C. Ge, and J. Jiang, “Towards extreme image compression with latent feature guidance and diffusion prior,”IEEE Transactions on Circuits and Systems for Video Technology, 2024

    2024

  9. [9]

    Rdeic: Accelerating diffusion-based extreme image compression with relay residual diffu- sion,

    Z. Li, Y . Zhou, H. Wei, C. Ge, and A. Mian, “Rdeic: Accelerating diffusion-based extreme image compression with relay residual diffu- sion,”IEEE Transactions on Circuits and Systems for Video Technology, 2025

    2025

  10. [10]

    Extremely low-bitrate image com- pression semantically disentangled by lmms from a human perception perspective,

    J. Song, L. Yang, and M. Feng, “Extremely low-bitrate image com- pression semantically disentangled by lmms from a human perception perspective,”arXiv preprint arXiv:2503.00399, 2025

    work page arXiv 2025

  11. [11]

    High- resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High- resolution image synthesis with latent diffusion models,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 10 684–10 695

    2022

  12. [12]

    Stablecodec: Taming one-step diffusion for extreme image compression,

    T. Zhang, X. Luo, L. Li, and D. Liu, “Stablecodec: Taming one-step diffusion for extreme image compression,”arXiv preprint arXiv:2506.21977, 2025

    work page arXiv 2025

  13. [13]

    Oscar: One-step diffusion codec across multiple bit-rates,

    J. Guo, Y . Ji, Z. Chen, K. Liu, M. Liu, W. Rao, W. Li, Y . Guo, and Y . Zhang, “Oscar: One-step diffusion codec across multiple bit-rates,” arXiv preprint arXiv:2505.16091, 2025

    work page arXiv 2025

  14. [14]

    Steering one-step diffusion model with fidelity-rich decoder for fast image compression,

    Z. Chen, M. Zhou, J. Guo, J. Yuan, Y . Ji, and Y . Zhang, “Steering one-step diffusion model with fidelity-rich decoder for fast image compression,”arXiv preprint arXiv:2508.04979, 2025

    work page arXiv 2025

  15. [15]

    Diffpc: Diffusion-based high perceptual fidelity image compression with semantic refinement,

    Y . Xia, Y . Zhou, J. Wang, B. An, H. Wang, Y . Wang, and B. Chen, “Diffpc: Diffusion-based high perceptual fidelity image compression with semantic refinement,” inThe Thirteenth International Conference on Learning Representations, 2025

    2025

  16. [16]

    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

    2022

  17. [17]

    T. M. Cover,Elements of information theory. John Wiley & Sons, 1999

    1999

  18. [18]

    End-to-end optimized image compression,

    J. Ball ´e, V . Laparra, and E. P. Simoncelli, “End-to-end optimized image compression,” inInternational Conference on Learning Representations, 2017

    2017

  19. [19]

    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, pp. 7939–7948

    2020

  20. [20]

    Enhanced invertible encoding for learned image compression,

    Y . Xie, K. L. Cheng, and Q. Chen, “Enhanced invertible encoding for learned image compression,” inProceedings of the 29th ACM international conference on multimedia, 2021, pp. 162–170

    2021

  21. [21]

    The devil is in the details: Window- based attention for image compression,

    R. Zou, C. Song, and Z. Zhang, “The devil is in the details: Window- based attention for image compression,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2022, pp. 17 492–17 501

    2022

  22. [22]

    Llic: Large receptive field transform coding with adaptive weights for learned image compression,

    W. Jiang, P. Ning, J. Yang, Y . Zhai, F. Gao, and R. Wang, “Llic: Large receptive field transform coding with adaptive weights for learned image compression,”IEEE Transactions on Multimedia, vol. 26, pp. 10 937– 10 951, 2024

    2024

  23. [23]

    Mambaic: State space models for high-performance learned image compression,

    F. Zeng, H. Tang, Y . Shao, S. Chen, L. Shao, and Y . Wang, “Mambaic: State space models for high-performance learned image compression,” inProceedings of the Computer Vision and Pattern Recognition Confer- ence, 2025, pp. 18 041–18 050

    2025

  24. [24]

    Linear attention modeling for learned image compression,

    D. Feng, Z. Cheng, S. Wang, R. Wu, H. Hu, G. Lu, and L. Song, “Linear attention modeling for learned image compression,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 7623–7632

    2025

  25. [25]

    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

    2018

  26. [26]

    Causal contextual prediction for learned image compression,

    Z. Guo, Z. Zhang, R. Feng, and Z. Chen, “Causal contextual prediction for learned image compression,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 32, no. 4, pp. 2329–2341, 2021

    2021

  27. [27]

    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

    2021

  28. [28]

    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

    2020

  29. [29]

    Mlic: Multi- reference entropy model for learned image compression,

    W. Jiang, J. Yang, Y . Zhai, P. Ning, F. Gao, and R. Wang, “Mlic: Multi- reference entropy model for learned image compression,” inProceedings of the 31st ACM International Conference on Multimedia, 2023, pp. 7618–7627

    2023

  30. [30]

    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

    2025

  31. [31]

    Learned image compression with dictionary-based entropy model,

    J. Lu, L. Zhang, X. Zhou, M. Li, W. Li, and S. Gu, “Learned image compression with dictionary-based entropy model,” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 12 850–12 859

    2025

  32. [32]

    Rethinking lossy compression: The rate- distortion-perception tradeoff,

    Y . Blau and T. Michaeli, “Rethinking lossy compression: The rate- distortion-perception tradeoff,” inInternational Conference on Machine Learning. PMLR, 2019, pp. 675–685

    2019

  33. [33]

    High- fidelity generative image compression,

    F. Mentzer, G. D. Toderici, M. Tschannen, and E. Agustsson, “High- fidelity generative image compression,”Advances in neural information processing systems, vol. 33, pp. 11 913–11 924, 2020

    2020

  34. [34]

    Multi-realism image compression with a conditional generator,

    E. Agustsson, D. Minnen, G. Toderici, and F. Mentzer, “Multi-realism image compression with a conditional generator,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 22 324–22 333

    2023

  35. [35]

    Neural image compression with text-guided encoding for both pixel-level and perceptual fidelity,

    H. Lee, M. Kim, J.-H. Kim, S. Kim, D. Oh, and J. Lee, “Neural image compression with text-guided encoding for both pixel-level and perceptual fidelity,”arXiv preprint arXiv:2403.02944, 2024

    work page arXiv 2024

  36. [36]

    A resid- ual diffusion model for high perceptual quality codec augmentation,

    N. F. Ghouse, J. Petersen, A. Wiggers, T. Xu, and G. Sautiere, “A resid- ual diffusion model for high perceptual quality codec augmentation,” arXiv preprint arXiv:2301.05489, 2023

    work page arXiv 2023

  37. [37]

    Consistency guided dif- fusion model with neural syntax for perceptual image compression,

    H. Kuang, Y . Ma, W. Yang, Z. Guo, and J. Liu, “Consistency guided dif- fusion model with neural syntax for perceptual image compression,” in Proceedings of the 32nd ACM International Conference on Multimedia, 2024, pp. 1622–1631

    2024

  38. [38]

    Controllable distortion-perception tradeoff through latent diffusion for neural image compression,

    C. Zhou, G. Lu, J. Li, X. Chen, Z. Cheng, L. Song, and W. Zhang, “Controllable distortion-perception tradeoff through latent diffusion for neural image compression,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 39, no. 10, 2025, pp. 10 725–10 733

    2025

  39. [39]

    Lossy image compression with conditional diffusion models,

    R. Yang and S. Mandt, “Lossy image compression with conditional diffusion models,”Advances in Neural Information Processing Systems, vol. 36, pp. 64 971–64 995, 2023

    2023

  40. [40]

    Correcting diffusion-based perceptual image compression with privileged end-to-end decoder,

    Y . Ma, W. Yang, and J. Liu, “Correcting diffusion-based perceptual image compression with privileged end-to-end decoder,” inProceedings of the 41st International Conference on Machine Learning, 2024, pp. 34 075–34 093

    2024

  41. [41]

    Multi-modality deep network for extreme learned image compression,

    X. Jiang, W. Tan, T. Tan, B. Yan, and L. Shen, “Multi-modality deep network for extreme learned image compression,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 37, no. 1, 2023, pp. 1033–1041

    2023

  42. [42]

    Towards image compression with perfect realism at ultra-low bitrates,

    M. Careil, M. J. Muckley, J. Verbeek, and S. Lathuili`ere, “Towards image compression with perfect realism at ultra-low bitrates,” inThe Twelfth International Conference on Learning Representations, 2023

    2023

  43. [43]

    Text+ sketch: Image compression at ultra low rates,

    E. Lei, Y . B. Uslu, H. Hassani, and S. S. Bidokhti, “Text+ sketch: Image compression at ultra low rates,”arXiv preprint arXiv:2307.01944, 2023

    work page arXiv 2023

  44. [44]

    Misc: Ultra-low bitrate image semantic compression driven by large multimodal model,

    C. Li, G. Lu, D. Feng, H. Wu, Z. Zhang, X. Liu, G. Zhai, W. Lin, and W. Zhang, “Misc: Ultra-low bitrate image semantic compression driven by large multimodal model,”IEEE Transactions on Image Processing, 2024

    2024

  45. [45]

    Linearly transformed color guide for low- bitrate diffusion based image compression,

    T. Bordin and T. Maugey, “Linearly transformed color guide for low- bitrate diffusion based image compression,”IEEE Transactions on Image Processing, 2024

    2024

  46. [46]

    Extreme image compression using fine-tuned vqgans,

    Q. Mao, T. Yang, Y . Zhang, Z. Wang, M. Wang, S. Wang, L. Jin, and S. Ma, “Extreme image compression using fine-tuned vqgans,” in2024 Data Compression Conference (DCC). IEEE, 2024, pp. 203–212

    2024

  47. [47]

    Taming transformers for high- resolution image synthesis,

    P. Esser, R. Rombach, and B. Ommer, “Taming transformers for high- resolution image synthesis,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 12 873–12 883

    2021

  48. [48]

    Toward extreme image rescaling with generative prior and invertible prior,

    H. Wei, C. Ge, Z. Li, X. Qiao, and P. Deng, “Toward extreme image rescaling with generative prior and invertible prior,”IEEE Transactions on Circuits and Systems for Video Technology, vol. 34, no. 7, pp. 6181– 6193, 2024

    2024

  49. [49]

    Hybridflow: Infusing continuity into masked codebook for extreme low-bitrate image compression,

    L. Lu, Y . Xie, W. Jiang, W. Wang, X. Lin, and Y . Wang, “Hybridflow: Infusing continuity into masked codebook for extreme low-bitrate image compression,” inProceedings of the 32nd ACM International Conference on Multimedia, 2024, pp. 3010–3018

    2024

  50. [50]

    Dlf: Extreme image compression with dual-generative latent fusion,

    N. Xue, Z. Jia, J. Li, B. Li, Y . Zhang, and Y . Lu, “Dlf: Extreme image compression with dual-generative latent fusion,”arXiv preprint arXiv:2503.01428, 2025. 11

    work page arXiv 2025

  51. [51]

    Lmm-driven semantic image-text coding for ultra low-bitrate learned image compression,

    S. Murai, H. Sun, and J. Katto, “Lmm-driven semantic image-text coding for ultra low-bitrate learned image compression,” in2024 IEEE Inter- national Conference on Visual Communications and Image Processing (VCIP). IEEE, 2024, pp. 1–5

    2024

  52. [52]

    Ultra lowrate image compression with semantic residual coding and compression-aware diffusion,

    A. Ke, X. Zhang, T. Chen, M. Lu, C. Zhou, J. Gu, and Z. Ma, “Ultra lowrate image compression with semantic residual coding and compression-aware diffusion,”arXiv preprint arXiv:2505.08281, 2025

    work page arXiv 2025

  53. [53]

    Diffo: Single-step diffusion for image compression at ultra-low bitrates,

    C. Park, J. C. Lee, and J. H. Ko, “Diffo: Single-step diffusion for image compression at ultra-low bitrates,”arXiv preprint arXiv:2506.16572, 2025

    work page arXiv 2025

  54. [54]

    Adversarial diffusion distillation,

    A. Sauer, D. Lorenz, A. Blattmann, and R. Rombach, “Adversarial diffusion distillation,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 87–103

    2024

  55. [55]

    Inceptionnext: When inception meets convnext,

    W. Yu, P. Zhou, S. Yan, and X. Wang, “Inceptionnext: When inception meets convnext,” inProceedings of the IEEE/cvf conference on computer vision and pattern recognition, 2024, pp. 5672–5683

    2024

  56. [56]

    Mambaout: Do we really need mamba for vision?

    W. Yu and X. Wang, “Mambaout: Do we really need mamba for vision?” inProceedings of the Computer Vision and Pattern Recognition Conference, 2025, pp. 4484–4496

    2025

  57. [57]

    Lsdir: A large scale dataset for image restoration,

    Y . Li, K. Zhang, J. Liang, J. Cao, C. Liu, R. Gong, Y . Zhang, H. Tang, Y . Liu, D. Demandolxet al., “Lsdir: A large scale dataset for image restoration,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 1775–1787

    2023

  58. [58]

    A unified end-to-end framework for efficient deep image compression,

    J. Liu, G. Lu, Z. Hu, and D. Xu, “A unified end-to-end framework for efficient deep image compression,”arXiv preprint arXiv:2002.03370, 2020

    work page arXiv 2002

  59. [59]

    Kodak photocd dataset,

    R. Franzen, “Kodak photocd dataset,” 1999. [Online]. Available: http://r0k.us/graphics/kodak/

    1999

  60. [60]

    Clic 2020: Challenge on learned image compression,

    G. Toderici, L. Theis, N. Johnston, E. Agustsson, F. Mentzer, J. Ball ´e, W. Shi, and R. Timofte, “Clic 2020: Challenge on learned image compression,” 2020

    2020

  61. [61]

    Testimages: a large-scale archive for testing visual devices and basic image processing algorithms

    N. Asuni and A. Giachetti, “Testimages: a large-scale archive for testing visual devices and basic image processing algorithms.” inSTAG, 2014, pp. 63–70

    2014

  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

    2018

  63. [63]

    Image quality assessment: Unifying structure and texture similarity,

    K. Ding, K. Ma, S. Wang, and E. P. Simoncelli, “Image quality assessment: Unifying structure and texture similarity,”IEEE transactions on pattern analysis and machine intelligence, vol. 44, no. 5, pp. 2567– 2581, 2020

    2020

  64. [64]

    Gans trained by a two time-scale update rule converge to a local nash equilibrium,

    M. Heusel, H. Ramsauer, T. Unterthiner, B. Nessler, and S. Hochreiter, “Gans trained by a two time-scale update rule converge to a local nash equilibrium,”Advances in neural information processing systems, vol. 30, 2017

    2017

  65. [65]

    Demystifying MMD GANs

    M. Bi ´nkowski, D. J. Sutherland, M. Arbel, and A. Gretton, “Demysti- fying mmd gans,”arXiv preprint arXiv:1801.01401, 2018

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  66. [66]

    Multiscale structural similarity for image quality assessment,

    Z. Wang, E. P. Simoncelli, and A. C. Bovik, “Multiscale structural similarity for image quality assessment,” inThe thrity-seventh asilomar conference on signals, systems & computers, 2003, vol. 2. Ieee, 2003, pp. 1398–1402

    2003

  67. [67]

    Lora: Low-rank adaptation of large language models

    E. J. Hu, Y . Shen, P. Wallis, Z. Allen-Zhu, Y . Li, S. Wang, L. Wang, W. Chenet al., “Lora: Low-rank adaptation of large language models.” ICLR, vol. 1, no. 2, p. 3, 2022

    2022

  68. [68]

    Dinov2: Learning robust visual features without supervision,

    M. Oquab, T. Darcet, T. Moutakanni, H. V o, M. Szafraniec, V . Khalidov, P. Fernandez, D. Haziza, F. Massa, A. El-Noubyet al., “Dinov2: Learning robust visual features without supervision,”Transactions on Machine Learning Research Journal, 2024

    2024

  69. [69]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clarket al., “Learning transferable visual models from natural language supervision,” inInternational conference on machine learning, 2021, pp. 8748–8763. Hao Weiis currently a Ph.D. candidate with the Institute of Artificial Intelligence and Robotics at...

    2021