pith. sign in

arxiv: 2504.04758 · v4 · submitted 2025-04-07 · 💻 cs.IT · math.IT

Feature Importance-Aware Deep Joint Source-Channel Coding for Computationally Efficient and Adjustable Image Transmission

Hansung Choi , Daewon Seo This is my paper

Pith reviewed 2026-05-22 21:15 UTC · model grok-4.3

classification 💻 cs.IT math.IT
keywords deepJSCCjoint source-channel codingimage transmissionfeature importancecomputational efficiencyadjustable complexityself-attention
0
0 comments X

The pith

FAJSCC reduces computation for image transmission while allowing independent encoder and decoder complexity control in one model.

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

The paper introduces a Feature Importance-Aware deepJSCC model called FAJSCC for transmitting images over communication channels. It targets the high computational demands of prior deep learning approaches that limit real-world use, while also enabling dynamic adaptation of resources. The design performs efficient operations separately along spatial and channel dimensions and applies self-attention only to selected important features whose number can be set independently at the encoder and decoder. Experiments across channel conditions show better reconstruction quality at lower overall cost than recent alternatives. The work also demonstrates for the first time that decoder processing of noisy features accounts for the largest share of computation.

Core claim

FAJSCC employs axis-dimension specialized computation to handle spatial and channel features efficiently and selective deformable self-attention on adaptively chosen important features to capture correlations with reduced cost. This yields superior image transmission performance under various channel conditions while using less computational complexity than state-of-the-art models. It is the first deepJSCC architecture that permits the number of selected important areas to be controlled separately by the encoder and the decoder within a single trained model.

What carries the argument

Axis-dimension specialized computation paired with selective deformable self-attention applied only to adaptively chosen important features, enabling both efficiency and separate control of encoder versus decoder complexity.

If this is right

  • Superior reconstruction quality is maintained across different channel conditions at lower overall computation than prior models.
  • Encoder and decoder computational budgets can be set independently after a single training run.
  • The largest computational demand arises from the decoder interpreting noisy received features.
  • Practical systems can allocate resources asymmetrically depending on device capabilities.

Where Pith is reading between the lines

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

  • The independent control finding could support asymmetric links where a low-power sender transmits to a high-resource receiver.
  • The decoder-cost observation suggests future designs should explore lighter decoder architectures or shared computation across multiple receivers.
  • The same selective-feature approach might apply to video or sensor data streams where content importance varies over time.

Load-bearing premise

That focusing computation on selected important features through axis-specialized operations and limited self-attention still preserves enough information for high-fidelity image reconstruction across changing channel conditions and image content.

What would settle it

An experiment that sets the number of selected important areas to a very low value in the decoder under poor channel conditions and checks whether reconstruction quality falls sharply compared with full selection.

Figures

Figures reproduced from arXiv: 2504.04758 by Daewon Seo, Hansung Choi.

Figure 2
Figure 2. Figure 2: FAJSCC Architecture. pConv LN LReLU Conv LReLU pConv LReLU pConv Conv Sigmoid Conv LReLU Avg Pooling pConv Sigmoid Channel Avg Window reshape pConv Softmax 𝑧𝑧𝑖𝑖𝑖𝑖 Spatial Information Spatial Importance Spatial Attention Window Importance Offset Channel Attention Root Processor Spatial Information Branch Spatial Importance Branch Spatial Attention Head Window Importance Head Offset Head Channel Attention He… view at source ↗
Figure 5
Figure 5. Figure 5: Procedures of selective deformable self-attention. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Efficiency comparison for various model sizes with respect to computational burden (GFLOPs) and model storage size (MB) for DIV [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: PSNR results under different channel and CPP environments for DIV [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: SSIM results under different channel and CPP environments for DIV [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: PSNR results of models trained at fixed SNR for DIV [PITH_FULL_IMAGE:figures/full_fig_p013_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: PSNR results of models trained at randomly sampled SNRs from [ [PITH_FULL_IMAGE:figures/full_fig_p013_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: PSNR results of models under the fast Rayleigh fading channel with estimated fading coefficients for DIV [PITH_FULL_IMAGE:figures/full_fig_p013_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Performance comparison for various importance ratios under different channel noises for DIV [PITH_FULL_IMAGE:figures/full_fig_p014_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: The first row shows the transmitted image at [PITH_FULL_IMAGE:figures/full_fig_p015_13.png] view at source ↗
read the original abstract

Recent advances in deep learning-based joint source-channel coding (deepJSCC) have substantially improved communication performance, but their high computational cost hinders practical deployment. Moreover, certain applications require the ability to dynamically adapt computational complexity. To address these issues, we propose a Feature Importance-Aware deepJSCC (FAJSCC) model for image transmission that is both computationally efficient and adjustable. FAJSCC employs axis-dimension specialized computation, which performs efficient operations individually for each spatial and channel axis, significantly reducing computational cost while representing features effectively. It further incorporates selective deformable self-attention, which applies self-attention only to selected and adaptively adjusted features, leveraging the importance and relations of input features to efficiently capture complex feature correlations. Another key feature of FAJSCC is that the number of selected important areas can be controlled separately by the encoder and the decoder, depending on the available computational budget. It makes FAJSCC the first deepJSCC architecture to allow independent adjustment of encoder and decoder complexity within a single trained model. Experimental results show that FAJSCC achieves superior image transmission performance under various channel conditions while requiring less computational complexity than recent state-of-the-art models. Furthermore, experiments independently varying the encoder and decoder's computational resources reveal, for the first time in the deepJSCC literature, that understanding the meaning of noisy features in the decoder demands the greatest computational cost. The code is publicly available at github.com/hansung-choi/FAJSCCv2.

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

0 major / 3 minor

Summary. The manuscript proposes Feature Importance-Aware deep Joint Source-Channel Coding (FAJSCC) for image transmission. It introduces axis-dimension specialized computation to reduce cost while preserving feature representation, combined with selective deformable self-attention applied only to adaptively chosen important features. A central feature is that the number of selected important areas can be controlled independently for the encoder and decoder at inference time within one trained model, claimed as the first such deepJSCC architecture. Experiments report superior rate-distortion performance under varying channel conditions with lower computational complexity than recent SOTA models, plus the observation that decoder processing of noisy features incurs the highest computational demand. Public code is released.

Significance. If the reported empirical gains and adjustability hold under the tested conditions, the work meaningfully advances practical deepJSCC deployment by simultaneously tackling efficiency and runtime adaptability—two key obstacles to real-world use. The public code and ablations directly support reproducibility of the efficiency and independent-control claims. The differential encoder/decoder complexity insight is a novel empirical contribution that could guide future architecture design.

minor comments (3)
  1. [§3] §3 (architecture description): the precise mechanism for independently selecting the number of important areas at encoder versus decoder (including how the selection mask is generated and transmitted) would benefit from an explicit equation or pseudocode block to make the inference-time adjustment fully reproducible from the text alone.
  2. [Table 2, Figure 4] Table 2 and Figure 4: the reported complexity metrics (FLOPs, parameters) should explicitly state whether they include or exclude the deformable attention overhead and the feature-selection module, as this directly affects the 'less computational complexity than SOTA' claim.
  3. [§4.3] §4.3 (ablation on independent budgets): the statement that 'understanding the meaning of noisy features in the decoder demands the greatest computational cost' is supported by the curves but would be strengthened by reporting the exact selection budgets (e.g., number of areas) used in each encoder/decoder configuration.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive summary, significance assessment, and recommendation of minor revision. The report accurately reflects the core contributions of FAJSCC regarding efficiency via axis-dimension specialized computation and selective deformable self-attention, as well as the novel independent encoder-decoder complexity control within a single model. No specific major comments are listed in the report.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper is an empirical ML architecture proposal. All central claims (superior rate-distortion performance, lower complexity, independent encoder/decoder budget control) rest on training, ablation studies, and test-set measurements under standard channel models. No derivation chain, uniqueness theorem, fitted-parameter prediction, or self-citation load-bearing step appears; the architecture description is internally consistent with the stated goals and the released code supplies direct falsifiable evidence. This is the normal non-circular outcome for a well-specified empirical contribution.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim depends on standard assumptions of deep neural network training and the effectiveness of the proposed modules for feature selection and attention; no new physical entities or ad-hoc constants beyond typical model hyperparameters are introduced.

free parameters (1)
  • Number of selected important areas
    This is chosen separately at encoder and decoder based on available computational budget rather than fixed during training.
axioms (1)
  • domain assumption Feature importance can be reliably estimated from input features to guide selective attention without losing critical information for reconstruction.
    Invoked in the design of selective deformable self-attention and the overall efficiency claims.

pith-pipeline@v0.9.0 · 5797 in / 1257 out tokens · 22845 ms · 2026-05-22T21:15:05.271672+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

65 extracted references · 65 canonical work pages · 1 internal anchor

  1. [1]

    On separation of source and channel coding in the finite block length regime,

    J. Ho, J. Meng, and E.-h. Yang, “On separation of source and channel coding in the finite block length regime,” inProc. Canadian Workshop Inf. Theory, Jun. 2013, pp. 92–95

    work page 2013

  2. [2]

    Multiple access channels with arbitrarily correlated sources,

    T. Cover, A. E. Gamal, and M. Salehi, “Multiple access channels with arbitrarily correlated sources,”IEEE Trans. Inf. Theory, vol. 26, no. 6, pp. 648–657, Nov. 1980

    work page 1980

  3. [3]

    Shannon-Kotel-Nikov mappings in joint source-channel coding,

    F. Hekland, P. A. Floor, and T. A. Ramstad, “Shannon-Kotel-Nikov mappings in joint source-channel coding,”IEEE Trans. Commun., vol. 57, no. 1, pp. 94–105, Jan. 2009

    work page 2009

  4. [4]

    Analog joint source-channel coding using non-linear curves and MMSE decoding,

    Y . Hu, J. Garcia-Frias, and M. Lamarca, “Analog joint source-channel coding using non-linear curves and MMSE decoding,”IEEE Trans. Commun., vol. 59, no. 11, pp. 3016–3026, Sep. 2011

    work page 2011

  5. [5]

    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. on Cogn. Commun. Netw., vol. 5, no. 3, pp. 567–579, May 2019

    work page 2019

  6. [6]

    Predictive and adaptive deep coding for wireless image transmission in semantic communication,

    W. Zhang, H. Zhang, H. Ma, H. Shao, N. Wang, and V . C. Leung, “Predictive and adaptive deep coding for wireless image transmission in semantic communication,”IEEE Trans. Wireless Commun., vol. 22, no. 8, pp. 5486–5501, Jan. 2023

    work page 2023

  7. [7]

    SwinJSCC: Taming swin transformer for deep joint source-channel coding,

    K. Yang, S. Wang, J. Dai, X. Qin, K. Niu, and P. Zhang, “SwinJSCC: Taming swin transformer for deep joint source-channel coding,”IEEE Trans. on Cogn. Commun. Netw., vol. 11, no. 1, pp. 90–104, Feb. 2025

    work page 2025

  8. [8]

    TCC-SemCom: A transformer-cnn complementary block based image semantic communication,

    G. Cheng, B. Chong, and H. Lu, “TCC-SemCom: A transformer-cnn complementary block based image semantic communication,”IEEE Commun. Lett., vol. 29, no. 3, pp. 625–629, Feb. 2025

    work page 2025

  9. [9]

    SNR-EQ-JSCC: Joint source-channel coding with snr-based embedding and query,

    H. Zhang and M. Tao, “SNR-EQ-JSCC: Joint source-channel coding with snr-based embedding and query,”IEEE Wireless Commun. Lett., vol. 14, no. 3, pp. 881–885, Jan. 2025

    work page 2025

  10. [10]

    Semantic importance-aware reordering-enhanced semantic communication system with OFDM transmission,

    Y . Liu, C. Dong, H. Liang, W. Li, Z. Bao, Z. Zheng, X. Xu, and P. Zhang, “Semantic importance-aware reordering-enhanced semantic communication system with OFDM transmission,”IEEE Internet Things J., vol. 12, no. 7, pp. 7938–7954, Apr. 2025

    work page 2025

  11. [11]

    Wireless image transmission using deep source channel coding with attention modules,

    J. Xu, B. Ai, W. Chen, A. Yang, P. Sun, and M. Rodrigues, “Wireless image transmission using deep source channel coding with attention modules,”IEEE Trans. Circuits Syst. Video Technol., vol. 32, no. 4, pp. 2315–2328, May 2021

    work page 2021

  12. [12]

    Mam- baJSCC: Adaptive deep joint source-channel coding with generalized state space model,

    T. Wu, Z. Chen, M. Tao, Y . Sun, X. Xu, W. Zhang, and P. Zhang, “Mam- baJSCC: Adaptive deep joint source-channel coding with generalized state space model,”arXiv preprint arXiv:2409.16592, Sep. 2024

    work page arXiv 2024

  13. [13]

    High perceptual quality wireless image delivery with denoising diffusion models,

    S. F. Yilmaz, X. Niu, B. Bai, W. Han, L. Deng, and D. G ¨und¨uz, “High perceptual quality wireless image delivery with denoising diffusion models,” inProc. IEEE Conf. Computer Commun. Workshop, May 2024, pp. 1–5

    work page 2024

  14. [14]

    Semantics- guided diffusion for deep joint source-channel coding in wireless image transmission,

    M. Zhang, H. Wu, G. Zhu, R. Jin, X. Chen, and D. G ¨und¨uz, “Semantics- guided diffusion for deep joint source-channel coding in wireless image transmission,”IEEE Trans. Wireless Commun., vol. 25, pp. 1547–1564, Jul. 2025, early access

    work page 2025

  15. [15]

    CDDM: Channel denoising diffusion models for wireless semantic communications,

    T. Wu, Z. Chen, D. He, L. Qian, Y . Xu, M. Tao, and W. Zhang, “CDDM: Channel denoising diffusion models for wireless semantic communications,”IEEE Trans. Wireless Commun., vol. 23, no. 9, pp. 11 168–11 183, Mar. 2024

    work page 2024

  16. [16]

    Bandwidth-agile image transmission with deep joint source-channel coding,

    D. B. Kurka and D. G ¨und¨uz, “Bandwidth-agile image transmission with deep joint source-channel coding,”IEEE Trans. Wireless Commun., vol. 20, no. 12, pp. 8081–8095, Jun. 2021

    work page 2021

  17. [17]

    DeepJSCC-1++: Robust and bandwidth-adaptive wireless image transmission,

    C. Bian, Y . Shao, and D. G ¨und¨uz, “DeepJSCC-1++: Robust and bandwidth-adaptive wireless image transmission,” inProc. IEEE Global Telecommun. Conf., Dec. 2023, pp. 3148–3154

    work page 2023

  18. [18]

    Deep joint source-channel coding for wireless image transmission with adaptive rate control,

    M. Yang and H.-S. Kim, “Deep joint source-channel coding for wireless image transmission with adaptive rate control,” inProc. IEEE Int. Conf. Acoust., Speech, Signal Process., May 2022, pp. 5193–5197

    work page 2022

  19. [19]

    Deep joint source-channel coding for adaptive image transmission over MIMO channels,

    H. Wu, Y . Shao, C. Bian, K. Mikolajczyk, and D. G ¨und¨uz, “Deep joint source-channel coding for adaptive image transmission over MIMO channels,”IEEE Trans. Wireless Commun., vol. 23, no. 10, pp. 15 002– 15 017, Jul. 2024

    work page 2024

  20. [20]

    Process- and-forward: Deep joint source-channel coding over cooperative relay networks,

    C. Bian, Y . Shao, H. Wu, E. Ozfatura, and D. G ¨und¨uz, “Process- and-forward: Deep joint source-channel coding over cooperative relay networks,”IEEE J. Sel. Areas Commun., vol. 43, no. 4, pp. 1118–1134, Jan. 2025

    work page 2025

  21. [21]

    Collaborative semantic communication for edge inference,

    W. F. Lo, N. Mital, H. Wu, and D. G ¨und¨uz, “Collaborative semantic communication for edge inference,”IEEE Wireless Commun. Lett., vol. 12, no. 7, pp. 1125–1129, Mar. 2023

    work page 2023

  22. [22]

    Importance-aware image segmentation-based semantic communication for autonomous driving,

    J. Lv, H. Tong, Q. Pan, Z. Zhang, X. He, T. Luo, and C. Yin, “Importance-aware image segmentation-based semantic communication for autonomous driving,”arXiv preprint arXiv:2401.10153, Jan. 2024

    work page arXiv 2024

  23. [23]

    Multi-task semantic communication with graph attention-based feature correlation extraction,

    X. Yu, T. Lv, W. Li, W. Ni, D. Niyato, and E. Hossain, “Multi-task semantic communication with graph attention-based feature correlation extraction,”IEEE Trans. Mobile Comput., vol. 24, no. 5, pp. 4371–4388, May 2025

    work page 2025

  24. [24]

    Rate-adaptable multitask- oriented semantic communication: An extended rate–distortion theory- based scheme,

    F. Liu, Z. Sun, Y . Yang, C. Guo, and S. Zhao, “Rate-adaptable multitask- oriented semantic communication: An extended rate–distortion theory- based scheme,”IEEE Internet Things J., vol. 11, no. 9, pp. 15 557–15 570, Jan. 2024

    work page 2024

  25. [25]

    Joint device-edge inference over wireless links with pruning,

    M. Jankowski, D. G ¨und¨uz, and K. Mikolajczyk, “Joint device-edge inference over wireless links with pruning,” inProc. IEEE Workshop Signal Process. Adv. Wireless Commun., May 2020, pp. 1–5

    work page 2020

  26. [26]

    Engineering a lightweight deep joint source-channel coding based semantic commu- nication system,

    W. Zhang, S. Wu, S. Meng, J. He, and Q. Zhang, “Engineering a lightweight deep joint source-channel coding based semantic commu- nication system,”IEEE Internet Things J., vol. 12, no. 1, pp. 458–471, Sep. 2024

    work page 2024

  27. [27]

    Lightweight deep joint source-channel coding for semantic communications over fading channels,

    W. Zhang, S. Wu, S. Meng, M. Liu, and Q. Zhang, “Lightweight deep joint source-channel coding for semantic communications over fading channels,” inProc. Int. Conf. Wireless Commun. Signal Process., Oct. 2024, pp. 1430–1435

    work page 2024

  28. [28]

    Low-rank decomposition for rate-adaptive deep joint source-channel coding,

    M. Xu, C.-T. Lam, Y . Liang, B. Ng, and S.-K. Im, “Low-rank decomposition for rate-adaptive deep joint source-channel coding,” in Proc. Int. Conf. Comput. Commun., Mar. 2022, pp. 58–64

    work page 2022

  29. [29]

    A novel lightweight joint source- channel coding design in semantic communications,

    X. Yu, D. Li, N. Zhang, and X. Shen, “A novel lightweight joint source- channel coding design in semantic communications,”IEEE Internet Things J., vol. 12, no. 11, pp. 18 447 – 18 450, Jun. 2025

    work page 2025

  30. [30]

    Multimodal- oriented interactive joint source-channel coding for lightweight semantic communication,

    X. Niu, L. Tan, J. Wu, W. Yuan, and T. Q. Quek, “Multimodal- oriented interactive joint source-channel coding for lightweight semantic communication,”IEEE Trans. V eh. Technol., vol. 74, no. 10, pp. 16 516– 16 520, Oct. 2025

    work page 2025

  31. [31]

    A progressive approach to joint source-channel coding for image super-resolution task in semantic communications,

    Z. Huang, Y . Jia, W. Wen, L. Liang, J. Yan, and N. Jiang, “A progressive approach to joint source-channel coding for image super-resolution task in semantic communications,”IEEE Wireless Commun. Lett., vol. 14, no. 7, pp. 2099–2103, Jul. 2025. 19

    work page 2099

  32. [32]

    Lightweight and robust wireless semantic communications,

    G. Chen, G. Nan, Z. Jiang, H. Du, R. Shi, Q. Cui, and X. Tao, “Lightweight and robust wireless semantic communications,”IEEE Commun. Lett., vol. 28, no. 11, pp. 2633–2637, Nov. 2024

    work page 2024

  33. [33]

    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

    work page 2023

  34. [34]

    A unified multi- task semantic communication system for multimodal data,

    G. Zhang, Q. Hu, Z. Qin, Y . Cai, G. Yu, and X. Tao, “A unified multi- task semantic communication system for multimodal data,”IEEE Trans. Commun., vol. 72, no. 7, pp. 4101–4116, Feb. 2024

    work page 2024

  35. [35]

    Outrageously large neural networks: The sparsely-gated mixture-of-experts layer,

    N. Shazeer, A. Mirhoseini, K. Maziarz, A. Davis, Q. Le, G. Hinton, and J. Dean, “Outrageously large neural networks: The sparsely-gated mixture-of-experts layer,” inProc. Int. Conf. Learn. Representations, Apr. 2017

    work page 2017

  36. [36]

    DynamicViT: Efficient vision transformers with dynamic token sparsification,

    Y . Rao, W. Zhao, B. Liu, J. Lu, J. Zhou, and C.-J. Hsieh, “DynamicViT: Efficient vision transformers with dynamic token sparsification,” inProc. Adv. Neural Inf. Process. Syst., vol. 34, Dec. 2021, pp. 13 937–13 949

    work page 2021

  37. [37]

    Branchynet: Fast inference via early exiting from deep neural networks,

    S. Teerapittayanon, B. McDanel, and H.-T. Kung, “Branchynet: Fast inference via early exiting from deep neural networks,” inProc. Int. Conf. Pattern Recognit., Dec. 2016, pp. 2464–2469

    work page 2016

  38. [38]

    Computation and transmission adaptive semantic communication for reliability-guarantee image recon- struction in IoT,

    C. Lin, Y . Guo, J. Hao, and Z. Zhang, “Computation and transmission adaptive semantic communication for reliability-guarantee image recon- struction in IoT,”Internet Things, vol. 28, no. 101383, pp. 1–14, Dec. 2024

    work page 2024

  39. [39]

    DD-JSCC: Dynamic deep joint source-channel coding for semantic communications,

    A. D. Raha, A. Adhikary, M. Gain, Y . Park, W. Saad, and C. S. Hong, “DD-JSCC: Dynamic deep joint source-channel coding for semantic communications,” inInd. Corporate Change, Jun. 2025, pp. 3754–3759

    work page 2025

  40. [40]

    CBAM: Convolutional block attention module,

    S. Woo, J. Park, J.-Y . Lee, and I. S. Kweon, “CBAM: Convolutional block attention module,” inProc. Eur . Conf. Comput. Vis., Sep. 2018, pp. 3–19

    work page 2018

  41. [41]

    Swin transformer: Hierarchical vision transformer using shifted windows,

    Z. Liu, Y . Lin, Y . Cao, H. Hu, Y . Wei, Z. Zhang, S. Lin, and B. Guo, “Swin transformer: Hierarchical vision transformer using shifted windows,” inProc. IEEE Conf. Comput. Vis. Pattern Recognit., Jun. 2021, pp. 10 012–10 022

    work page 2021

  42. [42]

    Semantic successive refinement: A generative ai-aided semantic communication framework,

    K. Zhang, L. Li, W. Lin, Y . Yan, R. Li, W. Cheng, and Z. Han, “Semantic successive refinement: A generative ai-aided semantic communication framework,”IEEE Trans. on Cogn. Commun. Netw., vol. 11, no. 2, pp. 687–699, Apr. 2025

    work page 2025

  43. [43]

    Task- scalable image semantic communication via conditional affine transforms and pixel-wise quality control,

    J. Wang, S. Yao, S. Wang, Z. Si, F. Wang, Z. Liu, and J. Dai, “Task- scalable image semantic communication via conditional affine transforms and pixel-wise quality control,” inProc. IEEE Wireless Commun. Netw. Conf., Mar. 2025

    work page 2025

  44. [44]

    Learned image compres- sion with discretized gaussian mixture likelihoods and attention modules,

    Z. Cheng, H. Sun, M. Takeuchi, and J. Katto, “Learned image compres- sion with discretized gaussian mixture likelihoods and attention modules,” inProc. IEEE Conf. Comput. Vis. Pattern Recognit., Jun. 2020, pp. 7939–7948

    work page 2020

  45. [45]

    ClassSR: A general framework to accelerate super-resolution networks by data characteristic,

    X. Kong, H. Zhao, Y . Qiao, and C. Dong, “ClassSR: A general framework to accelerate super-resolution networks by data characteristic,” inProc. IEEE Conf. Comput. Vis. Pattern Recognit., Jun. 2021, pp. 12 016–12 025

    work page 2021

  46. [46]

    Rate-distortion- complexity optimized framework for multi-model image compression,

    X. Hang, Z. Ge, H. Fan, C. Jia, S. Ma, and W. Gao, “Rate-distortion- complexity optimized framework for multi-model image compression,” IEEE Trans. Image Process., vol. 34, pp. 5385–5399, Aug. 2025

    work page 2025

  47. [47]

    Image quality assessment: from error visibility to structural similarity,

    Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image quality assessment: from error visibility to structural similarity,”IEEE Trans. Image Process., vol. 13, no. 4, pp. 600–612, Apr. 2004

    work page 2004

  48. [48]

    Deformable convolutional networks,

    J. Dai, H. Qi, Y . Xiong, Y . Li, G. Zhang, H. Hu, and Y . Wei, “Deformable convolutional networks,” inProc. IEEE Int. Conf. Computer Vision, Oct. 2017, pp. 764–773

    work page 2017

  49. [49]

    Efficient deformable convnets: Rethinking dynamic and sparse operator for vision applications,

    Y . Xiong, Z. Li, Y . Chen, F. Wang, X. Zhu, J. Luo, W. Wang, T. Lu, H. Li, Y . Qiaoet al., “Efficient deformable convnets: Rethinking dynamic and sparse operator for vision applications,” inProc. IEEE Conf. Comput. Vis. Pattern Recognit., Jun. 2024, pp. 5652–5661

    work page 2024

  50. [50]

    SCSC: A novel standards-compatible semantic communication frame- work for image transmission,

    X. Han, Y . Wu, Z. Gao, B. Feng, Y . Shi, D. G ¨und¨uz, and W. Zhang, “SCSC: A novel standards-compatible semantic communication frame- work for image transmission,”IEEE Trans. Commun., Jan. 2025

    work page 2025

  51. [51]

    Vision transformer with deformable attention,

    Z. Xia, X. Pan, S. Song, L. E. Li, and G. Huang, “Vision transformer with deformable attention,” inProc. IEEE Conf. Comput. Vis. Pattern Recognit., Jun. 2022, pp. 4794–4803

    work page 2022

  52. [52]

    CAMixerSR: Only details need more “attention

    Y . Wang, Y . Liu, S. Zhao, J. Li, and L. Zhang, “CAMixerSR: Only details need more “attention”,” inProc. IEEE Conf. Comput. Vis. Pattern Recognit., Jun. 2024, pp. 25 837–25 846

    work page 2024

  53. [53]

    Categorical reparameterization with Gumbel-Softmax,

    E. Jang, S. Gu, and B. Poole, “Categorical reparameterization with Gumbel-Softmax,” inProc. Int. Conf. Learn. Representations, Apr. 2017

    work page 2017

  54. [54]

    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., Jun. 2017, pp. 126–135

    work page 2017

  55. [55]

    Kodak PhotoCD dataset,

    “Kodak PhotoCD dataset,” http://r0k.us/graphics/kodak/, 1993

    work page 1993

  56. [56]

    Capacity and power allocation for fading mimo channels with channel estimation error,

    T. Yoo and A. Goldsmith, “Capacity and power allocation for fading mimo channels with channel estimation error,”IEEE Trans. Inf. Theory, vol. 52, no. 5, pp. 2203–2214, May 2006

    work page 2006

  57. [57]

    Deep learning-based scalable and robust channel estimator for wireless cellular networks,

    A. Lee, Y . Kwon, H. Park, and H. Lee, “Deep learning-based scalable and robust channel estimator for wireless cellular networks,”ETRI J., vol. 44, no. 6, pp. 915–924, Dec. 2022

    work page 2022

  58. [58]

    An Overview of Multi-Task Learning in Deep Neural Networks

    S. Ruder, “An overview of multi-task learning in deep neural networks,” arXiv preprint arXiv:1706.05098, Jun. 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  59. [59]

    Adversarial spatio-temporal learning for video deblurring,

    K. Zhang, W. Luo, Y . Zhong, L. Ma, W. Liu, and H. Li, “Adversarial spatio-temporal learning for video deblurring,”IEEE Trans. Image Process., vol. 28, no. 1, pp. 291–301, Jan. 2019

    work page 2019

  60. [60]

    MC-blur: A comprehensive benchmark for image deblurring,

    K. Zhang, T. Wang, W. Luo, W. Ren, B. Stenger, W. Liu, H. Li, and M.- H. Yang, “MC-blur: A comprehensive benchmark for image deblurring,” IEEE Trans. Circuits Syst. Video Technol., vol. 34, no. 5, pp. 3755–3767, May 2024

    work page 2024

  61. [61]

    Deep image deblurring: A survey,

    K. Zhang, W. Ren, W. Luo, W.-S. Lai, B. Stenger, M.-H. Yang, and H. Li, “Deep image deblurring: A survey,”Int. J. Comput. Vis., vol. 130, no. 9, pp. 2103–2130, Jun. 2022

    work page 2022

  62. [62]

    Survey of turbo, LDPC, and polar decoder ASIC implementations,

    S. Shao, P. Hailes, T.-Y . Wang, J.-Y . Wu, R. G. Maunder, B. M. Al- Hashimi, and L. Hanzo, “Survey of turbo, LDPC, and polar decoder ASIC implementations,”IEEE Commun. Surveys Tuts., vol. 21, no. 3, pp. 2309–2333, Jan. 2019

    work page 2019

  63. [63]

    CompressAI: a pytorch libra ry and evaluation platform for end -to-end compression research,

    J. B ´egaint, F. Racap ´e, S. Feltman, and A. Pushparaja, “Compressai: a pytorch library and evaluation platform for end-to-end compression research,”arXiv preprint arXiv:2011.03029, Nov. 2020

    work page arXiv 2011

  64. [64]

    Bjøntegaard delta (BD): A tutorial overview of the metric, evolution, challenges, and recommenda- tions,

    N. Barman, M. G. Martini, and Y . Reznik, “Bjøntegaard delta (BD): A tutorial overview of the metric, evolution, challenges, and recommenda- tions,”arXiv preprint arXiv:2401.04039, Jan. 2024

    work page arXiv 2024

  65. [65]

    Flickr30k entities: Collecting region-to-phrase correspondences for richer image-to-sentence models,

    B. A. Plummer, L. Wang, C. M. Cervantes, J. C. Caicedo, J. Hocken- maier, and S. Lazebnik, “Flickr30k entities: Collecting region-to-phrase correspondences for richer image-to-sentence models,” inProc. IEEE Int. Conf. Computer Vision, Dec. 2015, pp. 2641–2649

    work page 2015