pith. sign in

arxiv: 2606.06273 · v1 · pith:UDTPEZPOnew · submitted 2026-06-04 · 💻 cs.IT · cs.AI· math.IT

Adapting Diffusion Language Models for Lossless Pixel-Level Image Transmission

Tianqi Ren , Rongpeng Li , Xianfu Chen , Yingyu Li , Zhifeng Zhao This is my paper

Pith reviewed 2026-06-27 23:26 UTC · model grok-4.3

classification 💻 cs.IT cs.AImath.IT
keywords diffusion language modelslossless image transmissionseparate source-channel codingarithmetic codingdenoising scheduleHalton sequencetemperature calibrationexact recovery
0
0 comments X

The pith

Adapting discrete diffusion models enables synchronized arithmetic coding for exact pixel recovery over noisy channels.

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

The paper introduces DDM-SSCC, a separate source-channel coding scheme that repurposes a diffusion language model to restore pixel tokens progressively while performing arithmetic coding in reverse under bidirectional attention. This setup codes multiple masked tokens per denoising step and uses newly restored pixels as context for the next steps, producing a representation suited to noisy transmission. Three supporting modules—a Halton-guided order, a mask-ratio-aware cosine schedule, and temperature calibration—are added to align generation-style denoising with the exact probability tables needed for lossless arithmetic coding. Experiments across CIFAR10, DIV2K-LR-X4, and Kodak images transmitted over additive white Gaussian noise and Rayleigh fading channels report higher exact-recovery rates than both traditional lossless coders and semantic-communication baselines. The work therefore claims that diffusion-based progressive restoration can meet the dual demands of precise symbol modeling and channel robustness for pixel-perfect delivery.

Core claim

DDM-SSCC adapts a diffusion language model to pixel-token restoration and synchronized reverse arithmetic coding, allowing multiple masked tokens to be coded within one reverse denoising step; the progressive process supplies restored tokens as bidirectional context for subsequent steps, and the added Halton-guided order, mask-ratio-aware cosine schedule, and temperature calibration module together produce probability tables accurate enough for exact recovery, yielding higher exact-recovery rates than representative lossless and semantic baselines on CIFAR10, DIV2K-LR-X4, and Kodak over AWGN and Rayleigh channels.

What carries the argument

DDM-SSCC framework that adapts diffusion-language-model masked denoising to bidirectional arithmetic coding, using a Halton-guided denoising order, mask-ratio-aware cosine schedule, and temperature calibration to align generation probabilities with lossless coding requirements.

Load-bearing premise

The Halton-guided denoising order, mask-ratio-aware cosine schedule, and lightweight temperature calibration module successfully bridge the gap between masked denoising in diffusion models and the requirements of lossless arithmetic coding for exact pixel recovery.

What would settle it

Removing the three proposed modules and measuring whether exact-recovery rates on the same CIFAR10 and Kodak test sets over AWGN fall below those of standard raster-order autoregressive lossless coders would falsify the bridging claim.

Figures

Figures reproduced from arXiv: 2606.06273 by Rongpeng Li, Tianqi Ren, Xianfu Chen, Yingyu Li, Zhifeng Zhao.

Figure 1
Figure 1. Figure 1: Overall DDM-SSCC pipeline for lossless pixel-level image transmission. [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Comparison between raster-order autoregressive coding and discrete [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of the improved denoising strategy: (a) Halton-guided denoising position selection; (b) Cosine reverse denoising schedule; (c) Mask-ratio [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Visualization of remaining-mask entropy at the middle denoising stage [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Main performance comparison on CIFAR10, DIV2K-LR-X4 and Kodak under AWGN channels. The upper row reports PSNR and the lower row [PITH_FULL_IMAGE:figures/full_fig_p010_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Main performance comparison on CIFAR10, DIV2K-LR-X4 and Kodak under Rayleigh fading channels. The upper row reports PSNR and the lower [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Representative reconstruction example at SNR [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Representative reconstruction example at SNR [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗
Figure 11
Figure 11. Figure 11: Ablation study of the reverse denoising schedule on CIFAR10. [PITH_FULL_IMAGE:figures/full_fig_p012_11.png] view at source ↗
Figure 10
Figure 10. Figure 10: Ablation study of the denoising-position selection rule on CIFAR10. [PITH_FULL_IMAGE:figures/full_fig_p012_10.png] view at source ↗
read the original abstract

Lossless pixel-level image transmission is a fundamental regime beyond semantic communications, because exact recovery requires both accurate symbol probability modeling and reliable delivery over noisy channels. This paper proposes DDM-SSCC, a discrete-diffusion-model-based separate source-channel coding framework for lossless image transmission. Different from raster-order autoregressive coding, the proposed source codec adapts a diffusion language model to pixel-token restoration and performs synchronized reverse arithmetic coding under bidirectional attention, allowing multiple masked tokens to be coded within one reverse denoising step. This progressive restoration process also yields a more favorable source representation for noisy transmission, since newly restored tokens can serve as bidirectional context in subsequent denoising steps. To bridge the gap between generation-oriented masked denoising and lossless arithmetic coding, we further introduce a Halton-guided denoising order, a mask-ratio-aware cosine schedule, and a lightweight temperature calibration module. These designs respectively improve spatial coverage, adapt the denoising pace to context reliability, and calibrate the probability tables used by arithmetic coding. Experiments on CIFAR10, DIV2K-LR-X4, and Kodak over additive white Gaussian noise and Rayleigh fading channels show that DDM-SSCC achieves better exact-recovery performance than representative lossless and semantic communication baselines, while ablation studies verify the effectiveness of the proposed denoising order, schedule, and calibration modules.

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 paper proposes DDM-SSCC, a discrete-diffusion-model-based separate source-channel coding (SSCC) framework for lossless pixel-level image transmission. It adapts a diffusion language model to perform synchronized reverse arithmetic coding under bidirectional attention, replacing raster-order autoregressive modeling with masked denoising that restores multiple tokens per step. Three adaptations—Halton-guided denoising order, mask-ratio-aware cosine schedule, and lightweight temperature calibration—are introduced to align generation-oriented diffusion with the determinism required for exact arithmetic decoding. Experiments on CIFAR10, DIV2K-LR-X4, and Kodak over AWGN and Rayleigh fading channels report superior exact-recovery performance versus representative lossless and semantic baselines, with ablations supporting the three modules.

Significance. If the probability tables remain consistent and reversible across synchronized encoder/decoder steps, the approach could meaningfully extend diffusion models beyond generation to exact-recovery transmission tasks, leveraging bidirectional context for more robust source representations. The explicit ablation studies on the denoising order, schedule, and calibration constitute a strength in isolating component contributions.

major comments (2)
  1. [§3 and §4] §3 (Method) and §4 (Experiments): The central claim of exact pixel recovery on perfect channels requires that the Halton order, mask-ratio-aware cosine schedule, and temperature calibration produce identical probability tables at corresponding encoder and decoder steps. No formal argument, pseudocode verification, or empirical check (e.g., zero-error reconstruction rate on noiseless channels) is supplied to confirm that these modules preserve the determinism needed for arithmetic decoding; the skeptic concern therefore remains unaddressed and is load-bearing for the lossless guarantee.
  2. [Table 1 and §4.2] Table 1 and §4.2: The reported exact-recovery gains over baselines are presented without accompanying bit-rate tables, per-image error breakdowns, or statistical significance tests. This makes it impossible to determine whether the improvements are driven by the diffusion adaptations or by differences in effective rate allocation, undermining the cross-dataset and cross-channel claims.
minor comments (2)
  1. [§2] §2 (Related Work): The comparison to prior diffusion-based compression works omits recent discrete diffusion language model papers that also target token-level probability modeling; adding these would strengthen positioning.
  2. [Figure 3] Figure 3: The visualization of the Halton-guided order would benefit from an explicit overlay of the corresponding mask-ratio schedule to illustrate their joint effect on spatial coverage.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment point-by-point below, indicating where revisions will be made to strengthen the presentation and validation of our claims.

read point-by-point responses

  1. Referee: [§3 and §4] §3 (Method) and §4 (Experiments): The central claim of exact pixel recovery on perfect channels requires that the Halton order, mask-ratio-aware cosine schedule, and temperature calibration produce identical probability tables at corresponding encoder and decoder steps. No formal argument, pseudocode verification, or empirical check (e.g., zero-error reconstruction rate on noiseless channels) is supplied to confirm that these modules preserve the determinism needed for arithmetic decoding; the skeptic concern therefore remains unaddressed and is load-bearing for the lossless guarantee.

    Authors: We agree that an explicit verification of determinism is essential to support the lossless guarantee. Section 3 describes how the Halton-guided order, mask-ratio-aware schedule, and temperature calibration are designed to maintain consistent and reversible probability tables across synchronized steps, but we acknowledge the absence of a direct empirical check on noiseless channels. In the revised manuscript, we will add results in §4 showing 100% exact-recovery rates on perfect channels (zero-error reconstruction) for all datasets to empirically confirm this property. revision: yes

  2. Referee: [Table 1 and §4.2] Table 1 and §4.2: The reported exact-recovery gains over baselines are presented without accompanying bit-rate tables, per-image error breakdowns, or statistical significance tests. This makes it impossible to determine whether the improvements are driven by the diffusion adaptations or by differences in effective rate allocation, undermining the cross-dataset and cross-channel claims.

    Authors: We acknowledge that additional details on rates and variability would help isolate the contributions of our adaptations from rate allocation effects. The current experiments emphasize exact-recovery performance under the SSCC framework, where source rates are determined by the arithmetic coding with the learned probabilities. In the revision, we will augment Table 1 and §4.2 with bit-rate tables, per-image error breakdowns, and statistical significance tests (e.g., paired t-tests) to address this concern. revision: yes

Circularity Check

0 steps flagged

No significant circularity; independent adaptations validated experimentally

full rationale

The paper proposes DDM-SSCC by adapting a diffusion language model with three explicitly introduced components (Halton-guided order, mask-ratio-aware cosine schedule, temperature calibration) to enable synchronized bidirectional arithmetic coding. These modules are presented as new designs to address the generation-vs-determinism mismatch, with effectiveness shown via ablation studies and comparative experiments on CIFAR10/DIV2K/Kodak rather than any reduction to fitted inputs, self-definitional equations, or load-bearing self-citations. The central claim of exact-recovery superiority rests on measured performance, not on any quantity that equals its own construction by definition. No steps match the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

Since only the abstract is available, the ledger is based on components explicitly mentioned; no specific numerical values or additional entities are detailed.

free parameters (2)
  • cosine schedule parameters
    Mask-ratio-aware cosine schedule to adapt the denoising pace to context reliability.
  • temperature calibration parameters
    Lightweight module for calibrating probability tables in arithmetic coding.
axioms (1)
  • domain assumption Diffusion models can be synchronized with reverse arithmetic coding under bidirectional attention for multiple tokens.
    Fundamental to the source codec adaptation.

pith-pipeline@v0.9.1-grok · 5764 in / 1096 out tokens · 33744 ms · 2026-06-27T23:26:21.195332+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

52 extracted references · 1 linked inside Pith

  1. [1]

    Deep joint source- channel coding for wireless image transmission,

    E. Bourtsoulatze, D. B. Kurka, and D. G ¨und¨uz, “Deep joint source- channel coding for wireless image transmission,”IEEE Trans. Cogn. Commun. Netw., vol. 5, no. 3, pp. 567–579, Sep. 2019

    2019

  2. [2]

    Alternate learning- based SNR-adaptive sparse semantic visual transmission,

    S. Tong, X. Yu, R. Li, K. Lu, Z. Zhao, and H. Zhang, “Alternate learning- based SNR-adaptive sparse semantic visual transmission,”IEEE Trans. Wireless Commun., vol. 24, no. 2, pp. 1737–1752, Feb. 2025

    2025

  3. [3]

    DeepMA: End-to-end deep multiple access for wireless image transmission in semantic communication,

    W. Zhang, K. Bai, S. Zeadally, H. Zhang, H. Shao, H. Ma, and V . C. M. Leung, “DeepMA: End-to-end deep multiple access for wireless image transmission in semantic communication,”IEEE Trans. Cogn. Commun. Netw., vol. 10, no. 2, pp. 387–402, Apr. 2024

    2024

  4. [4]

    Lightweight joint source-channel coding for semantic communications,

    Y . Jia, Z. Huang, K. Luo, and W. Wen, “Lightweight joint source-channel coding for semantic communications,”IEEE Commun. Lett., vol. 27, no. 12, pp. 3161–3165, Dec. 2023

    2023

  5. [5]

    Language modeling is compression,

    G. Del ´etang, A. Ruoss, P.-A. Duquenne, E. Catt, T. Genewein, C. Mat- tern, J. Grau-Moya, L. K. Wenliang, M. Aitchison, L. Orseau, M. Hutter, and J. Veness, “Language modeling is compression,” inProc. Int. Conf. Learn. Represent. (ICLR), Vienna, Austria, May 2024

    2024

  6. [6]

    Compression represents intelligence linearly,

    Y . Huang, J. Zhang, Z. Shan, and J. He, “Compression represents intelligence linearly,” Apr. 2024, arXiv preprint arXiv:2404.09937. 13

    arXiv 2024

  7. [7]

    Lossless data compression by large models,

    Z. Li, C. Huang, X. Wang, H. Hu, C. Wyeth, D. Bu, Q. Yu, W. Gao, X. Liu, and M. Li, “Lossless data compression by large models,”Nat. Mach. Intell., vol. 7, no. 5, pp. 794–799, May 2025

    2025

  8. [8]

    Large language model for lossless image compression with visual prompts,

    J. Du, C. Zhou, N. Cao, G. Chen, Y . Chen, Z. Cheng, L. Song, G. Lu, and W. Zhang, “Large language model for lossless image compression with visual prompts,” Feb. 2025, arXiv preprint arXiv:2502.16163

    arXiv 2025

  9. [9]

    Large language models for lossless image compression: Next-pixel prediction in language space is all you need,

    K. Chen, P. Zhang, H. Liu, J. Liu, Y . Liu, J. Huang, S. Wang, H. Yan, and H. Li, “Large language models for lossless image compression: Next-pixel prediction in language space is all you need,” inProc. Adv. Neural Inf. Process. Syst. (NeurIPS), San Diego, USA, Dec. 2025

    2025

  10. [10]

    Separate source channel coding is still what you need: An LLM-based rethinking,

    T. Ren, R. Li, M. min Zhao, X. Chen, G. Liu, Y . Yang, Z. Zhao, and H. Zhang, “Separate source channel coding is still what you need: An LLM-based rethinking,”ZTE Commun., vol. 23, no. 1, pp. 30–44, Mar. 2025

    2025

  11. [11]

    Error correction code transformer,

    Y . Choukroun and L. Wolf, “Error correction code transformer,” inProc. Adv. Neural Inf. Process. Syst. (NeurIPS), New Orleans, USA, Nov. 2022, pp. 38 695–38 705

    2022

  12. [12]

    A foundation model for error correction codes,

    ——, “A foundation model for error correction codes,” inProc. Int. Conf. Learn. Represent. (ICLR), Vienna, Austria, May 2024

    2024

  13. [13]

    U-shaped error correction code transform- ers,

    D.-T. Nguyen and S. Kim, “U-shaped error correction code transform- ers,”IEEE Trans. Cogn. Commun. Netw., vol. 11, no. 2, pp. 975–986, Apr. 2025

    2025

  14. [14]

    Self- critical alternate learning-based semantic broadcast communication,

    Z. Lu, R. Li, M. Lei, C. Wang, Z. Zhao, and H. Zhang, “Self- critical alternate learning-based semantic broadcast communication,” IEEE Trans. Commun., vol. 73, no. 5, pp. 3347–3361, May 2025

    2025

  15. [15]

    Semantic communications using foundation models: Design approaches and open issues,

    P. Jiang, C.-K. Wen, X. Yi, X. Li, S. Jin, and J. Zhang, “Semantic communications using foundation models: Design approaches and open issues,”IEEE Wireless Commun., vol. 31, no. 3, pp. 76–84, Jun. 2024

    2024

  16. [16]

    Large AI model-based semantic communications,

    F. Jiang, Y . Peng, L. Dong, K. Wang, K. Yang, C. Pan, and X. You, “Large AI model-based semantic communications,”IEEE Wireless Com- mun., vol. 31, no. 3, pp. 68–75, Jun. 2024

    2024

  17. [17]

    Generative AI-driven semantic communication networks: Architecture, technologies and applications,

    C. Liang, H. Du, Y . Sun, D. Niyato, J. Kang, D. Zhao, and M. A. Imran, “Generative AI-driven semantic communication networks: Architecture, technologies and applications,”IEEE Trans. Cogn. Commun. Netw., vol. 11, no. 1, pp. 27–47, Feb. 2025

    2025

  18. [18]

    Large AI model empowered multimodal semantic communications,

    F. Jiang, L. Dong, Y . Peng, K. Wang, K. Yang, C. Pan, and X. You, “Large AI model empowered multimodal semantic communications,” IEEE Commun. Mag., vol. 63, no. 1, pp. 76–82, Jan. 2025

    2025

  19. [19]

    DiT-JSCC: Rethinking deep JSCC with diffusion transformers and semantic representations,

    K. Tan, J. Dai, S. Wang, G. Lu, S. Shao, K. Niu, W. Zhang, and P. Zhang, “DiT-JSCC: Rethinking deep JSCC with diffusion transformers and semantic representations,” 2026, arXiv preprint arXiv:2601.03112

    Pith/arXiv arXiv 2026

  20. [20]

    Low-density parity-check codes,

    R. G. Gallager, “Low-density parity-check codes,”IRE Trans. Inf. Theory, vol. 8, no. 1, pp. 21–28, Jan. 1962

    1962

  21. [21]

    Channel polarization: A method for constructing capacity- achieving codes for symmetric binary-input memoryless channels,

    E. Arikan, “Channel polarization: A method for constructing capacity- achieving codes for symmetric binary-input memoryless channels,”IEEE Trans. Inf. Theory, vol. 55, no. 7, pp. 3051–3073, Jul. 2009

    2009

  22. [22]

    Deep learning methods for improved decoding of linear codes,

    E. Nachmani, E. Marciano, L. Lugosch, W. J. Gross, D. Burshtein, and Y . Be’ery, “Deep learning methods for improved decoding of linear codes,”IEEE J. Sel. Top. Signal Process., vol. 12, no. 1, pp. 119–131, Feb. 2018

    2018

  23. [23]

    Context-based, adaptive, lossless image coding,

    X. Wu and N. Memon, “Context-based, adaptive, lossless image coding,” IEEE Trans. Commun., vol. 45, no. 4, pp. 437–444, Apr. 1997

    1997

  24. [24]

    The LOCO-I lossless image compression algorithm: Principles and standardization into JPEG- LS,

    M. J. Weinberger, G. Seroussi, and G. Sapiro, “The LOCO-I lossless image compression algorithm: Principles and standardization into JPEG- LS,”IEEE Trans. Image Process., vol. 9, no. 8, pp. 1309–1324, Aug. 2000

    2000

  25. [25]

    JPEG XL next-generation image com- pression architecture and coding tools,

    J. Alakuijala, R. van Asseldonk, S. Boukortt, M. Bruse, I.-M. Comsa, M. Firsching, T. Fischbacher, S. Gomez, E. Kliuchnikov, R. Obryk, K. Potempa, A. Rhatushnyak, J. Sneyers, Z. Szabadka, L. Vandevenne, L. Versari, and J. Wassenberg, “JPEG XL next-generation image com- pression architecture and coding tools,” inProc. SPIE Appl. Digit. Image Process. XLII,...

    2019

  26. [26]

    Practical full resolution learned lossless image compression,

    F. Mentzer, E. Agustsson, M. Tschannen, R. Timofte, and L. V . Gool, “Practical full resolution learned lossless image compression,” inProc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), Long Beach, USA, Jun. 2019, pp. 10 629–10 638

    2019

  27. [27]

    Integer discrete flows and lossless compression,

    E. Hoogeboom, J. W. T. Peters, R. van den Berg, and M. Welling, “Integer discrete flows and lossless compression,” inProc. Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 32, Vancouver, Canada, Dec. 2019, pp. 12 134–12 144

    2019

  28. [28]

    PILC: Practical image lossless compression with an end-to-end GPU oriented neural framework,

    N. Kang, S. Qiu, S. Zhang, Z. Li, and S.-T. Xia, “PILC: Practical image lossless compression with an end-to-end GPU oriented neural framework,” inProc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), New Orleans, USA, Jun. 2022, pp. 3739–3748

    2022

  29. [29]

    Learned lossless image compression based on bit plane slicing,

    Z. Zhang, H. Wang, Z. Chen, and S. Liu, “Learned lossless image compression based on bit plane slicing,” inProc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), Seattle, USA, Jun. 2024, pp. 27 579–27 588

    2024

  30. [30]

    CALLIC: Content adaptive learning for lossless image compression,

    D. Li, Y . Bai, K. Wang, J. Jiang, X. Liu, and W. Gao, “CALLIC: Content adaptive learning for lossless image compression,” inProc. AAAI Conf. Artif. Intell., vol. 39, no. 5, Philadelphia, USA, Feb. 2025, pp. 4679– 4688

    2025

  31. [31]

    Deep lossy plus residual coding for lossless and near-lossless image compression,

    Y . Bai, X. Liu, K. Wang, X. Ji, X. Wu, and W. Gao, “Deep lossy plus residual coding for lossless and near-lossless image compression,”IEEE Trans. Pattern Anal. Mach. Intell., vol. 46, no. 5, pp. 3577–3594, May 2024

    2024

  32. [32]

    Fitted neural lossless image compres- sion,

    Z. Zhang, Z. Chen, and S. Liu, “Fitted neural lossless image compres- sion,” inProc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit. (CVPR), Nashville, USA, Jun. 2025, pp. 23 249–23 258

    2025

  33. [33]

    Generalized Kraft inequality and arithmetic coding,

    J. J. Rissanen, “Generalized Kraft inequality and arithmetic coding,” IBM J. Res. Dev., vol. 20, no. 3, pp. 198–203, May 1976

    1976

  34. [34]

    Arithmetic coding for data compression,

    I. H. Witten, R. M. Neal, and J. G. Cleary, “Arithmetic coding for data compression,”Commun. ACM, vol. 30, no. 6, pp. 520–540, Jun. 1987

    1987

  35. [35]

    Arithmetic coding for data compression,

    P. G. Howard and J. S. Vitter, “Arithmetic coding for data compression,” Proc. IEEE, vol. 82, no. 6, pp. 857–865, Jun. 1994

    1994

  36. [36]

    LLMZip: Lossless text compression using large language models,

    C. S. K. Valmeekam, K. Narayanan, D. Kalathil, J.-F. Chamberland, and S. Shakkottai, “LLMZip: Lossless text compression using large language models,” Jun. 2023, arXiv preprint arXiv:2306.04050

    arXiv 2023

  37. [37]

    Generative pretraining from pixels,

    M. Chen, A. Radford, R. Child, J. Wu, H. Jun, D. Luan, and I. Sutskever, “Generative pretraining from pixels,” inProc. Int. Conf. Mach. Learn. (ICML), ser. Proc. Mach. Learn. Res., vol. 119, Virtual Edition, Jul. 2020, pp. 1691–1703

    2020

  38. [38]

    Structured denoising diffusion models in discrete state-spaces,

    J. Austin, D. D. Johnson, J. Ho, D. Tarlow, and R. van den Berg, “Structured denoising diffusion models in discrete state-spaces,” inProc. Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 34, Virtual Edition, Dec. 2021, pp. 17 981–17 993

    2021

  39. [39]

    Diffusion-LM improves controllable text generation,

    X. L. Li, J. Thickstun, I. Gulrajani, P. Liang, and T. B. Hashimoto, “Diffusion-LM improves controllable text generation,” inProc. Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 35, New Orleans, USA, Dec. 2022, pp. 4328–4343

    2022

  40. [40]

    Simple and effective masked diffusion language models,

    S. S. Sahoo, M. Arriola, Y . Schiff, A. Gokaslan, E. Marroquin, J. T. Chiu, A. M. Rush, and V . Kuleshov, “Simple and effective masked diffusion language models,” inProc. Adv. Neural Inf. Process. Syst. (NeurIPS), vol. 37, Vancouver, Canada, Dec. 2024

    2024

  41. [41]

    Block diffusion: Interpolating between autoregressive and diffusion language models,

    M. Arriola, A. Gokaslan, J. T. Chiu, Z. Yang, Z. Qi, J. Han, S. S. Sahoo, and V . Kuleshov, “Block diffusion: Interpolating between autoregressive and diffusion language models,” inProc. Int. Conf. Learn. Represent. (ICLR), Singapore, Apr. 2025

    2025

  42. [42]

    Large language diffusion models,

    S. Nie, F. Zhu, Z. You, X. Zhang, J. Ou, J. Hu, J. Zhou, Y . Lin, J.- R. Wen, and C. Li, “Large language diffusion models,” inProc. Adv. Neural Inf. Process. Syst. (NeurIPS), San Diego, USA, Dec. 2025

    2025

  43. [43]

    Scaling diffusion language models via adaptation from autoregressive models,

    S. Gong, S. Agarwal, Y . Zhang, J. Ye, L. Zheng, M. Li, C. An, P. Zhao, W. Bi, H. Peng, J. Han, and L. Kong, “Scaling diffusion language models via adaptation from autoregressive models,” inProc. Int. Conf. Learn. Represent. (ICLR), Singapore, Apr. 2025

    2025

  44. [44]

    MaskGIT: Masked generative image transformer,

    H. Chang, H. Zhang, L. Jiang, C. Liu, and W. T. Freeman, “MaskGIT: Masked generative image transformer,” inProc. IEEE/CVF Conf. Com- put. Vis. Pattern Recognit. (CVPR), New Orleans, USA, Jun. 2022, pp. 11 315–11 325

    2022

  45. [45]

    Halton scheduler for masked generative image transformer,

    V . Besnier, M. Chen, D. Hurych, E. Valle, and M. Cord, “Halton scheduler for masked generative image transformer,” inProc. Int. Conf. Learn. Represent. (ICLR), Singapore, Apr. 2025

    2025

  46. [46]

    T. M. Cover and J. A. Thomas,Elements of Information Theory, 2nd ed. Hoboken, NJ, USA: John Wiley & Sons, 2006

    2006

  47. [47]

    Algorithm 247: Radical-inverse quasi- random point sequence,

    J. H. Halton and G. B. Smith, “Algorithm 247: Radical-inverse quasi- random point sequence,”Commun. ACM, vol. 7, no. 12, pp. 701–702, Dec. 1964

    1964

  48. [48]

    On calibration of modern neural networks,

    C. Guo, G. Pleiss, Y . Sun, and K. Q. Weinberger, “On calibration of modern neural networks,” inProc. Int. Conf. Mach. Learn. (ICML), ser. Proc. Mach. Learn. Res., vol. 70, Sydney, Australia, Aug. 2017, pp. 1321–1330

    2017

  49. [49]

    Learning multiple layers of features from tiny images,

    A. Krizhevsky, “Learning multiple layers of features from tiny images,” University of Toronto, Toronto, ON, Canada, Tech. Rep., 2009

    2009

  50. [50]

    NTIRE 2017 challenge on single image super-resolution: Dataset and study,

    E. Agustsson and R. Timofte, “NTIRE 2017 challenge on single image super-resolution: Dataset and study,” inProc. IEEE Conf. Comput. Vis. Pattern Recognit. Workshops (CVPRW), Honolulu, USA, Jul. 2017, pp. 1122–1131

    2017

  51. [51]

    Kodak lossless true color image suite,

    R. Franzen, “Kodak lossless true color image suite,” Available online: http://r0k.us/graphics/kodak/, 1999

    1999

  52. [52]

    Block- based adaptive vector lifting schemes for multichannel image coding,

    A. Benazza-Benyahia, J.-C. Pesquet, and H. Masmoudi, “Block- based adaptive vector lifting schemes for multichannel image coding,” EURASIP J. Image Video Process., vol. 2007, pp. 1–12, 2007

    2007