pith. sign in

arxiv: 2604.19515 · v1 · submitted 2026-04-21 · 💻 cs.IT · math.IT

Constructive Approaches to Perception-Aware Lossy Source Coding: Information-Theoretic Guidelines

Ali Hussein , Jun Chen , Chao Tian , S. Sandeep Pradhan This is my paper

Pith reviewed 2026-05-10 01:21 UTC · model grok-4.3

classification 💻 cs.IT math.IT
keywords perception-aware source codingrate-distortion-perceptioninformation theoretic guidelineslossy compressionconstructive codingunit circle examplecommon randomness
0
0 comments X

The pith

Rate-distortion-perception theory supplies guidelines for designing practical perception-aware lossy coders.

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

The paper surveys different information-theoretic formulations of rate-distortion-perception problems and distills them into practical guidelines for constructing perception-aware coding systems. It employs a unit-circle source as a pedagogical example to demonstrate these principles in both one-shot and asymptotic settings. Practitioners would care because this approach provides theoretical insight for system design rather than treating limits only as benchmarks or relying exclusively on the expressive power of neural networks. The work also addresses the role of common randomness and connections to standard lossy coding.

Core claim

By surveying rate-distortion-perception theory, the authors show that its principles can be turned into concrete design guidelines for implementable perception-aware lossy source coding schemes, illustrated in detail by the unit-circle example that unifies one-shot and asymptotic views while clarifying common randomness and universal representations.

What carries the argument

The rate-distortion-perception formulations distilled into guidelines, with the unit-circle example serving as the illustrative mechanism for architectural principles and tradeoffs.

If this is right

  • Implementable coding schemes can be developed by applying the distilled guidelines from the theory.
  • Common randomness is necessary for achieving certain perception levels in the schemes.
  • Universal representations can be identified that support multiple perception constraints.
  • Perception-aware coding connects to conventional lossy coding in specific ways that inform when extra constraints are needed.

Where Pith is reading between the lines

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

  • These guidelines could be applied to guide the architecture of neural network based codecs for images or video.
  • Testing the guidelines on real-world sources might reveal additional principles not captured by the unit-circle model.
  • Future work could derive similar guidelines for other distortion and perception measures beyond the surveyed ones.

Load-bearing premise

The surveyed formulations of rate-distortion-perception and the unit-circle example sufficiently represent the key principles that apply to general practical coding systems.

What would settle it

A concrete falsifier would be if a coding scheme constructed according to the guidelines performs no better than or worse than a black-box neural network design in terms of the rate-distortion-perception tradeoff on a standard source like Gaussian or image data.

Figures

Figures reproduced from arXiv: 2604.19515 by Ali Hussein, Chao Tian, Jun Chen, S. Sandeep Pradhan.

Figure 1
Figure 1. Figure 1: Illustration of the supports of X˜, X ′ , and Xˆ (with P = 0.04) in the unit-circle example for 1-bit quantization. Specifically, X˜ is uniformly distributed over the two points (± 2 π , 0) ≈ (±0.637, 0), X ′ is uniformly distributed over the unit circle centered at the origin, and Xˆ is uniformly distributed over two semicircles of radius 1 − π 5 √ π2−4 ≈ 0.741, centered at (± 2 5 √ π2−4 , 0) ≈ (±0.165, 0… view at source ↗
Figure 2
Figure 2. Figure 2: Due to the perception constraint, pXˆ must lie within the Wasserstein-2 ball of radius √ P centered at pZ . Among all such pZˆ, the one that minimizes W2(pZˆ, pZ˜) is located at the point where the geodesic from pZ to pZ˜ intersects the boundary of this ball. For this choice of pZˆ, we have W2(pZˆ, pZ˜) = W2(pZ , pZ˜) − √ P. with Z˜ := E[Z|W]. Moreover, the lower bound in (9) is attained by the following i… view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of the supports of Xˆ across different values of P in the unit-circle example for 1-bit quantization. Note that Xˆ coincides with X ′ when P = 0, and with X˜ when P → ∞ (more precisely, when P ≥ 1 − 4 π2 ≈ 0.595). It is instructive to interpret the unit circle as an abstract manifold of natural images. In the 1-bit quantization scenario, minimizing the mean squared error requires the reconstru… view at source ↗
Figure 4
Figure 4. Figure 4: Illustration of the distinction between the computations of [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Illustration of the supports of X˜, X ′ , and Xˆ (with P = 0.04) in the unit-circle example for 1-bit quantization with unlimited common randomness. In this case, X˜, X ′ , and Xˆ are uniformly distributed over circles centered at the origin with radii 2 π ≈ 0.637, 1, and 0.8, respectively, and are deterministically related through scaling. In the large-N limit, X˜, X′ , and Xˆ are uniformly distributed ov… view at source ↗
Figure 6
Figure 6. Figure 6: Comparison of the distortion–perception tradeoffs in the unit-circle example for [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Comparison of the supports of X˜, X ′ , and X˜ (with P = 0.04) in the unit-circle example for 1-bit quantization with 1-bit of common randomness (M = 2, N = 2) and 2-bit quantization with no common randomness (M = 4, N = 1). Specifically, in the first setting, X˜ is uniformly distributed over the four points (± 2 π , 0) ≈ (±0.637, 0) and (0, ± 2 π ) ≈ (0, ±0.637), while Xˆ is uniformly distributed over fou… view at source ↗
Figure 8
Figure 8. Figure 8: Comparison of the distortion-perception tradeoffs in the unit-circle example for [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: System diagram of perception-aware lossy source coding. [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Illustration of a lattice-based construction. The blue points represent the intermediate lattice [PITH_FULL_IMAGE:figures/full_fig_p027_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Consider a random vector Z in R k with E[∥Z∥ 2 ] < ∞, and a representation W generated according to some conditional distribution pW|Z . Let Z˜ be the MMSE estimate of Z based on W, and let Zˆ be a reconstruction derived from W. For a given perception measure Ξ, the constraint Ξ(pZ , pZˆ) ≤ P defines a perceptually permissible set (illustrated as an elliptical region) within which pZˆ must lie. The interp… view at source ↗
read the original abstract

Perception-aware lossy source coding has attracted significant recent interest. It augments the classical distortion criterion with an explicit perception constraint, thereby enabling more refined control over fidelity and perceptual quality. Despite rapid progress, the diversity of rate-distortion-perception formulations and their underlying assumptions remains poorly understood by many practitioners. In particular, there is often a tendency to rely heavily on the expressive power of deep neural networks and generative models without clear theoretical guidance, using fundamental limits merely as performance benchmarks rather than as sources of design insight. This tutorial paper aims to bridge this gap by surveying information-theoretic principles that can be leveraged to develop constructive approaches to perception-aware lossy source coding. We distill practical guidelines implied by rate-distortion-perception theory and demonstrate how they inform the design of implementable coding schemes. A simple unit-circle example is used as a pedagogical tool to illustrate key ideas, architectural principles, and tradeoffs in an intuitive and unified manner. Both one-shot and asymptotic settings are examined to highlight conceptual similarities and operational differences. We also clarify the role of common randomness and the notion of universal representation, and elucidate the connections between perception-aware and conventional lossy source coding. Overall, this tutorial provides a principled foundation for developing perception-aware compression systems that go beyond black-box model design.

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. This tutorial surveys rate-distortion-perception (RDP) formulations from information theory, distills them into practical design guidelines for perception-aware lossy source coding, and illustrates the guidelines via a unit-circle pedagogical example in both one-shot and asymptotic regimes. It emphasizes the roles of common randomness and universal representations while clarifying connections to classical lossy coding, aiming to move practitioners beyond black-box neural-network designs.

Significance. If the distilled guidelines accurately reflect the underlying RDP mathematics and the unit-circle example successfully conveys transferable architectural principles (e.g., when common randomness is required or how perception constraints alter rate-distortion trade-offs), the paper would offer a valuable pedagogical resource that helps bridge theory and constructive implementation in a field dominated by empirical deep-learning approaches.

major comments (2)
  1. The central claim that RDP principles 'distill into practical guidelines' whose implications transfer to implementable schemes rests on the unit-circle example; however, the example's low-dimensional rotational symmetry and simple distortion/perception functionals leave open whether the same principles survive non-convex high-dimensional optimization, learned perceptual metrics, or finite-blocklength regimes that dominate practical systems. A dedicated subsection should explicitly delineate which lessons are expected to generalize and which are artifacts of the pedagogical setup.
  2. The abstract states that both one-shot and asymptotic settings are examined to highlight 'conceptual similarities and operational differences,' yet without explicit comparison of the resulting guidelines (e.g., how the role of common randomness changes across regimes), it is unclear whether the distilled design rules are regime-specific or unified.
minor comments (1)
  1. The abstract refers to 'the diversity of rate-distortion-perception formulations' but does not list the specific formulations surveyed; an early table or enumerated list would improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which help clarify the scope of our pedagogical example and the presentation of regime-specific insights. We address each major comment below and will revise the manuscript to incorporate the suggested clarifications.

read point-by-point responses

  1. Referee: The central claim that RDP principles 'distill into practical guidelines' whose implications transfer to implementable schemes rests on the unit-circle example; however, the example's low-dimensional rotational symmetry and simple distortion/perception functionals leave open whether the same principles survive non-convex high-dimensional optimization, learned perceptual metrics, or finite-blocklength regimes that dominate practical systems. A dedicated subsection should explicitly delineate which lessons are expected to generalize and which are artifacts of the pedagogical setup.

    Authors: We agree that the unit-circle example is deliberately simplified for pedagogical clarity, leveraging rotational symmetry to illustrate core RDP concepts such as the role of common randomness in achieving optimal perception-distortion trade-offs and the distinction between distortion and perception constraints. The underlying information-theoretic results surveyed in the paper (e.g., from the RDP formulations in the literature) are dimension-agnostic and apply to general settings. However, we acknowledge that specific numerical trade-offs in the example may not directly carry over to non-convex high-dimensional cases or learned metrics. To address this, we will add a dedicated subsection that explicitly delineates expected generalizations (e.g., the necessity of common randomness for certain perception levels, as derived from the theory) versus setup-specific artifacts (e.g., closed-form solutions due to symmetry). This subsection will also discuss how the guidelines can inform practical designs in more complex regimes, referencing connections to finite-blocklength analyses where relevant. revision: yes

  2. Referee: The abstract states that both one-shot and asymptotic settings are examined to highlight 'conceptual similarities and operational differences,' yet without explicit comparison of the resulting guidelines (e.g., how the role of common randomness changes across regimes), it is unclear whether the distilled design rules are regime-specific or unified.

    Authors: We appreciate this point on presentation. The manuscript already examines both regimes to highlight similarities (e.g., common randomness enabling better perception-distortion trade-offs) and differences (e.g., asymptotic achievability vs. one-shot constraints). However, we agree that an explicit side-by-side comparison of the distilled guidelines would enhance clarity and demonstrate whether the rules are unified or regime-specific. We will revise the paper by adding a dedicated comparison subsection (or expanded discussion) that directly contrasts the guidelines across regimes, with particular emphasis on how the role of common randomness and universal representations evolves or remains consistent between one-shot and asymptotic settings. revision: yes

Circularity Check

0 steps flagged

No circularity: tutorial survey with pedagogical example

full rationale

The paper is a tutorial surveying existing rate-distortion-perception formulations from the literature and distilling guidelines for constructive coding schemes. It employs a unit-circle example purely as an illustrative pedagogical device to show conceptual similarities between one-shot and asymptotic settings, the role of common randomness, and connections to conventional lossy coding. No load-bearing derivation, prediction, or uniqueness claim reduces by construction to a fitted parameter, self-definition, or self-citation chain; the central content consists of exposition and unification of prior independent results. Standard self-citations of foundational RDP work are present but non-circular, as they reference externally established theory rather than serving as the sole justification for new claims within this manuscript.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The tutorial rests on standard information-theoretic axioms such as the definition of rate-distortion-perception functions and the operational meaning of common randomness; no new free parameters or invented entities are introduced in the abstract.

axioms (1)
  • domain assumption Rate-distortion-perception functions are well-defined and admit operational interpretations in both one-shot and asymptotic regimes.
    Invoked when distilling guidelines from the theory for constructive schemes.

pith-pipeline@v0.9.0 · 5532 in / 1185 out tokens · 21468 ms · 2026-05-10T01:21:15.636243+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

47 extracted references · 47 canonical work pages

  1. [1]

    The perception-distortion tradeoff,

    Y . Blau and T. Michaeli, “The perception-distortion tradeoff,” inProc. IEEE Conf. Comp. Vision and Pattern Recog. (CVPR), 2018, pp. 6288–6237

    work page 2018

  2. [2]

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

    work page 2006

  3. [3]

    Distribution preserving quantization with dithering and transformation,

    M. Li, J. Klejsa, and W. B. Kleijn, “Distribution preserving quantization with dithering and transformation,” IEEE Signal Process. Lett., vol. 17, no. 12, pp. 1014–1017, Dec. 2010

    work page 2010

  4. [4]

    Multiple description distribution preserving quantization,

    J. Klejsa, G. Zhang, M. Li, and W. B. Kleijn, “Multiple description distribution preserving quantization,” IEEE Trans. Signal Process., vol. 61, no. 24, pp. 6410–6422, Dec. 2013

    work page 2013

  5. [5]

    Randomized quantization and source coding with constrained output distribution,

    N. Saldi, T. Linder, and S. Y ¨uksel, “Randomized quantization and source coding with constrained output distribution,” IEEE Trans. Inf. Theory, vol. 61, no. 1, pp. 91–106, Jan. 2015

    work page 2015

  6. [6]

    Output constrained lossy source coding with limited common randomness,

    N. Saldi, T. Linder, and S. Y ¨uksel, “Output constrained lossy source coding with limited common randomness,” IEEE Trans. Inf. Theory, vol. 61, no. 9, pp. 4984–4998, Sep. 2015

    work page 2015

  7. [7]

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

    Y . Blau and T. Michaeli, “Rethinking lossy compression: The rate-distortion-perception tradeoff,” in Proc. ACM Int. Conf. Mach. Learn. (ICML), 2019, pp. 675–685

    work page 2019

  8. [8]

    Introducing the perception-distortion tradeoff into the rate-distortion theory of general information sources,

    R. Matsumoto, “Introducing the perception-distortion tradeoff into the rate-distortion theory of general information sources,” IEICE Comm. Express, vol. 7, no. 11, pp. 427–431, 2018

    work page 2018

  9. [9]

    Rate-distortion-perception tradeoff of variable-length source coding for general information sources,

    R. Matsumoto, “Rate-distortion-perception tradeoff of variable-length source coding for general information sources,” IEICE Comm. Express, vol. 8, no. 2, pp. 38–42, 2019

    work page 2019

  10. [10]

    On perceptual lossy compression: The cost of perceptual reconstruction and an optimal training framework,

    Z. Yan, F. Wen, R. Ying, C. Ma, and P. Liu, “On perceptual lossy compression: The cost of perceptual reconstruction and an optimal training framework,” in Proc. ACM Int. Conf. Mach. Learn. (ICML), 2021, pp. 11682–11692

    work page 2021

  11. [11]

    A coding theorem for the rate-distortion-perception function,

    L. Theis and A. B. Wagner, “A coding theorem for the rate-distortion-perception function,” in Proc. Neural Compress. Workshop Int. Conf. Learn. Represent. (ICLR), 2021, pp. 1–5

    work page 2021

  12. [12]

    On the advantages of stochastic encoders,

    L. Theis and E. Agustsson, “On the advantages of stochastic encoders,” in Proc. Neural Compress. Workshop Int. Conf. Learn. Represent. (ICLR), 2021, pp. 1–8

    work page 2021

  13. [13]

    Universal rate-distortion-perception representations for lossy compression,

    G. Zhang, J. Qian, J. Chen, and A. Khisti, “Universal rate-distortion-perception representations for lossy compression,” in Proc. Adv. Neural Inf. Process. Syst. (NeurIPS), 2021, pp. 11517–11529

    work page 2021

  14. [14]

    A rate-distortion-perception theory for binary sources,

    J. Qian, G. Zhang, J. Chen, and A. Khisti, “A rate-distortion-perception theory for binary sources,” in Proc. Int. Zurich Seminar Inf. Commun. (IZS), 2022, pp. 34–38

    work page 2022

  15. [15]

    Lossy compression with distribution shift as entropy constrained optimal transport,

    H. Liu, G. Zhang, J. Chen, A. Khisti, “Lossy compression with distribution shift as entropy constrained optimal transport,” in Proc. Int. Conf. Learn. Represent. (ICLR), 2022, pp. 1–34

    work page 2022

  16. [16]

    Optimally controllable perceptual lossy compression,

    Z. Yan, F. Wen, and P. Liu, “Optimally controllable perceptual lossy compression,” in Proc. ACM Int. Conf. Mach. Learn. (ICML), 2022, pp. 24911–24928. 31

    work page 2022

  17. [17]

    Cross-domain lossy compression as entropy constrained optimal transport,

    H. Liu, G. Zhang, J. Chen and A. Khisti, “Cross-domain lossy compression as entropy constrained optimal transport,” IEEE J. Sel. Areas Inf. Theory, vol. 3, no. 3, pp. 513–527, Sep. 2022

    work page 2022

  18. [18]

    On the rate-distortion-perception function,

    J. Chen, L. Yu, J. Wang, W. Shi, Y . Ge, and W. Tong, “On the rate-distortion-perception function,” IEEE J. Sel. Areas Inf. Theory, vol. 3, no. 4, pp. 664–673, Dec. 2022

    work page 2022

  19. [19]

    The rate-distortion-perception tradeoff: The role of common randomness,

    A. B. Wagner, “The rate-distortion-perception tradeoff: The role of common randomness,” 2022, arXiv:2202.04147. [Online] Available: https://arxiv.org/abs/2202.04147

    work page arXiv 2022

  20. [20]

    The rate-distortion-perception trade-off with side information,

    Y . Hamdi and D. G ¨und¨uz, “The rate-distortion-perception trade-off with side information,” in Proc. IEEE Int. Symp. Inf. Theory (ISIT), 2023, pp. 1056–1061

    work page 2023

  21. [21]

    Conditional rate-distortion-perception trade-off,

    X. Niu, D. G ¨und¨uz, B. Bai, and W. Han, “Conditional rate-distortion-perception trade-off,” in Proc. IEEE Int. Symp. Inf. Theory (ISIT), 2023, pp. 1068–1073

    work page 2023

  22. [22]

    On the choice of perception loss function for learned video compression,

    S. Salehkalaibar, B. Phan, J. Chen, W. Yu, and A. Khisti, “On the choice of perception loss function for learned video compression,” in Proc. Adv. Neural Inf. Process. Syst. (NeurIPS), 2023, pp. 1–19

    work page 2023

  23. [23]

    On the computation of the Gaussian rate-distortion-perception function,

    G. Serra, P. A. Stavrou, and M. Kountouris, “On the computation of the Gaussian rate-distortion-perception function,” IEEE J. Sel. Areas Inf. Theory, vol. 5, pp. 314–330, 2023

    work page 2023

  24. [24]

    The rate-distortion-perception trade-off: The role of private randomness,

    Y . Hamdi, A. B. Wagner, and D. Gund ¨uz, “The rate-distortion-perception trade-off: The role of private randomness,” in Proc. IEEE Int. Symp. Inf. Theory (ISIT), 2024, pp. 1083–1088

    work page 2024

  25. [25]

    Staggered quantizers for perfect perceptual quality: A connection between quantizers with common randomness and without,

    R. Zhou and C. Tian, “Staggered quantizers for perfect perceptual quality: A connection between quantizers with common randomness and without,” Proc. IEEE Int. Symp. Inf. Theory Workshops (ISIT-W), 2024, pp. 1–6

    work page 2024

  26. [26]

    Rate-limited quantum-to-classical optimal transport in finite and continuous-variable quantum systems,

    H. M. Garmaroudi, S. Sandeep Pradhan and J. Chen, “Rate-limited quantum-to-classical optimal transport in finite and continuous-variable quantum systems,” IEEE Trans. Inf. Theory, vol. 70, no. 11, pp. 7892–7922, Nov. 2024

    work page 2024

  27. [27]

    Rate-distortion-perception tradeoff based on the conditional-distribution perception measure,

    S. Salehkalaibar, J. Chen, A. Khisti, and W. Yu, “Rate-distortion-perception tradeoff based on the conditional-distribution perception measure,” IEEE Trans. Inf. Theory, vol. 70, no. 12, pp. 8432–8454, Dec. 2024

    work page 2024

  28. [28]

    Output-constrained lossy source coding with application to rate-distortion-perception theory,

    L. Xie, L. Li, J. Chen, and Z. Zhang, “Output-constrained lossy source coding with application to rate-distortion-perception theory,” IEEE Trans. Commun., vol. 73, no. 3, pp. 1801–1815, Mar. 2025

    work page 2025

  29. [29]

    A constrained Talagrand’s transportation inequality with application to rate-distortion-perception theory,

    L. Xie, L. Li, J. Chen, L. Yu, and Z. Zhang, “A constrained Talagrand’s transportation inequality with application to rate-distortion-perception theory,” Entropy, vol. 27, pp. 1–13, Apr. 2025

    work page 2025

  30. [30]

    Rate-distortion-perception tradeoff for Gaussian vector sources,

    J. Qian, S. Salehkalaibar, J. Chen, A. Khisti, W. Yu, W. Shi, Y . Ge, and W. Tong, “Rate-distortion-perception tradeoff for Gaussian vector sources,” IEEE J. Sel. Areas Inf. Theory, vol. 6, pp. 1–17, 2025

    work page 2025

  31. [31]

    Rate-distortion-perception theory for the quadratic Wasserstein space,

    X. Qu, J. Chen, L. Yu, and X. Xu, “Rate-distortion-perception theory for the quadratic Wasserstein space,” IEEE Trans. Inf. Theory, vol. 71, no. 11, pp. 8247–8261, Nov. 2025

    work page 2025

  32. [32]

    Gaussian rate-distortion-perception coding and entropy-constrained scalar quantization,

    L. Xie, L. Li, J. Chen, L. Yu, and Z. Zhang, “Gaussian rate-distortion-perception coding and entropy-constrained scalar quantization,” IEEE Trans. Commun., vol. 74, pp. 3298–3312, 2026

    work page 2026

  33. [33]

    J. Chen, Y . Fang, A. Khisti, A. ¨Ozg¨ur and N. Shlezinger, ”Information compression in the AI era: Recent advances and future challenges,” IEEE J. Sel. Areas Commun., vol. 43, no. 7, pp. 2333–2348, Jul. 2025

    work page 2025

  34. [34]

    A theory of the distortion-perception tradeoff in Wasserstein space,

    D. Freirich, T. Michaeli, and R. Meir, “A theory of the distortion-perception tradeoff in Wasserstein space,” in Proc. Adv. Neural Inf. Process. Syst. (NeurIPS), 2021, pp. 25661–25672

    work page 2021

  35. [35]

    On universal quantization,

    J. Ziv, “On universal quantization,” IEEE Trans. Inf. Theory, vol. 31, no. 3, pp. 344–347, May 1985

    work page 1985

  36. [36]

    Posterior-mean rectified flow: Towards minimum MSE photo-realistic image restoration,

    G. Ohayon, T. Michaeli, and M. Elad “Posterior-mean rectified flow: Towards minimum MSE photo-realistic image restoration,” in Proc. Int. Conf. Learn. Represent. (ICLR), 2025, pp. 1–40

    work page 2025

  37. [37]

    Flow straight and fast: Learning to generate and transfer data with rectified flow,

    X. Liu, C. Gong, and Q. Liu, “Flow straight and fast: Learning to generate and transfer data with rectified flow,” in Proc. Int. Conf. Learn. Represent. (ICLR), 2023, pp. 1–33. 32

    work page 2023

  38. [38]

    Characterization of the distortion-perception tradeoff for finite channels with arbitrary metrics,

    D. Freirich, N. Weinberger and R. Meir, “Characterization of the distortion-perception tradeoff for finite channels with arbitrary metrics,” Proc. IEEE Int. Symp. Inf. Theory (ISIT), 2024, pp. 238–243

    work page 2024

  39. [39]

    End-to-end optimized image compression

    J. Ball ´e, V . Laparra, and E. P. Simoncelli, “End-to-end optimized image compression.” inProc. Int. Conf. Learn. Represent. (ICLR), 2017, pp. 1–27

    work page 2017

  40. [40]

    Joint autoregressive and hierarchical priors for learned image compression,

    D. Minnen, J. Ball ´e, G. D. Toderici, “Joint autoregressive and hierarchical priors for learned image compression,” in Proc. Adv. Neural Inf. Process. Syst. (NeurIPS), 2018, pp. 10771–10780

    work page 2018

  41. [41]

    Distributed channel synthesis,

    P. Cuff, “Distributed channel synthesis,” IEEE Trans. Inf. Theory, vol. 59, no. 11, pp. 7071–7096, Nov. 2013

    work page 2013

  42. [42]

    Strong functional representation lemma and applications to coding theorems,

    C. T. Li and A. E. Gamal, “Strong functional representation lemma and applications to coding theorems,” IEEE Trans. Inf. Theory, vol. 64, no. 11, pp. 6967–6978, Nov. 2018

    work page 2018

  43. [43]

    Lossy quantum source coding with a global error criterion based on a posterior reference map,

    T. A. Atif, M. A. Sohail and S. S. Pradhan, “Lossy quantum source coding with a global error criterion based on a posterior reference map,” IEEE Trans. Inf. Theory, vol. 70, no. 5, pp. 3470–3498, May 2024

    work page 2024

  44. [44]

    Distributed source coding using Abelian group codes: A new achievable rate-distortion region,

    D. Krithivasan and S. S. Pradhan, “Distributed source coding using Abelian group codes: A new achievable rate-distortion region,” IEEE Trans. Inf. Theory, vol. 57, no. 3, pp. 1495–1519, Mar. 2011

    work page 2011

  45. [45]

    Optimal neural compressors for the rate-distortion-perception tradeoff,

    E. Lei, H. Hassani, and S. Saeedi Bidokhti, “Optimal neural compressors for the rate-distortion-perception tradeoff,” in Proc. Adv. Neural Inf. Process. Syst. (NeurIPS), 2025, pp. –

    work page 2025

  46. [46]

    Wasserstein distortion: Unifying fidelity and realism,

    Y . Qiu, A. B. Wagner, J. Balle, and L. Theis, “Wasserstein distortion: Unifying fidelity and realism,” in Proc. 58th Annu. Conf. Inf. Sci. Syst. (CISS), 2024, pp. 1–6

    work page 2024

  47. [47]

    Revisiting rate–distortion–perception theory: A new perspective,

    J. Chen, “Revisiting rate–distortion–perception theory: A new perspective,” preprint

    work page