Pith. sign in

REVIEW 5 major objections 5 minor 55 references

Semantic-Aware Visual Information Transmission With Key Information Extraction Over Wireless Networks

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper claims that a foreground/background split with a shared background library beats direct deep JSCC on PSNR and cuts transmitted data by over 20 times.

desk verdict Reasonable engineering idea, but the evaluation doesn't support the headline claims: the PSNR comparison is not resource-matched and the mask equations are inverted. read the letter →

arxiv 2506.12786 v1 pith:YRJGQAM7 submitted 2025-06-15 cs.CV

Chen Zhu , Kang Liang , Jianrong Bao , Zhouxiang Zhao , Zhaohui Yang , Zhaoyang Zhang , Mohammad Shikh-Bahaei This is my paper · ORCID
classification cs.CV
keywords semanticcommunicationdeepjointsource-channelcoding(JSCC)wirelessimagetransmissionkeyinformationextractionbackgroundmatchingsynthesis6Gnetworksresource-constrained
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that a semantic-aware wireless image transmission scheme can beat a plain end-to-end learned codec by not sending what the receiver can already know. It separates each image into a foreground subject and background, transmits only the foreground through a deep joint source-channel coding (JSCC) model, and at the receiver stitches the foreground onto the best-matching background from a shared library. The authors report higher peak signal-to-noise ratio than direct deep JSCC under the same channel conditions, with the largest margin at low SNR, and claim more than 20 times reduction in transmitted data relative to the original image. This matters because it points toward a way to push high-resolution image and video through resource-limited 6G links without requiring heavy GPU processing on the transmitter side. The gains, however, are conditional on a reusable background being available and on clean foreground extraction, which the paper itself treats as the fallback case.

What carries the argument

The load-bearing mechanism is the join of three components: (1) a foreground/background decomposition step in which a pose detector builds a convex-hull mask over the person and a segmentation tool crops out the foreground region; (2) a background library with a matching step that compares the masked image against masked candidates using binary feature descriptors and Hamming distance, choosing the library background with the highest number of matches above a threshold; and (3) an adaptive-rate deep joint source-channel coding (JSCC) network, whose policy network generates a binary mask to keep only active latent features, giving a coding-stage compression ratio of one third. The decomposition means only the small foreground crop is passed through the encoder, so the JSCC model spends its channel budget on the salient content rather than on background texture. At the receiver, the background is synthesized locally from the matched library entry, restoring the full image. The same decomposition also feeds the optional dynamic-background and frame-interpolation extensions for video.

What would settle it

The paper's claims would be settled by measuring, under identical SNR, the PSNR of its pipeline versus direct deep JSCC on a set of images whose backgrounds are deliberately absent from the library, and by recording the fraction of image area occupied by the foreground when the 20x compression figure is quoted. If the pipeline does not beat direct transmission in the no-match fallback case, or if the compression gain scales with foreground area rather than holding at 20x for all matching scenes, the central claim of universal improvement would be false.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central discovery is that foreground/background decomposition converts redundant image content into a lookup operation. Instead of encoding the whole frame, the system masks out the human subject using pose detection, masks the same region out of each candidate library background, compares the masked images by feature descriptors and Hamming distance, and selects the library background with the most matches. The cropped foreground region is then the only thing that enters an adaptive-rate deep JSCC encoder; after transmission, the receiver restores the crop to its original size and composites it onto the matched background. The authors claim this yields higher PSNR than sending the full image through the same JSCC model, especially when the channel SNR is low, and that the payload shrinks by more than 20 times for typical video-conference, surveillance, and remote-streaming scenes. The discovery extends to video by assembling a dynamic background library from the least-occluded pixels across frames, and to base-station scheduling by a resource-allocation rule that decides per user whether to transmit directly or via key-information extraction.

Load-bearing premise

The method's gains depend on a good match for the background being present in the shared library and on the foreground being cleanly separable; when the library has no suitable background, the paper lets the image be transmitted directly, and the claimed PSNR and 20x compression advantages do not apply.

Editorial extensions

If this is right

  • In scenes where a background match exists, the proposed pipeline reports higher PSNR than direct deep JSCC transmission at the same SNR, with larger gains at low SNR.
  • Transmitted data volume is cut by more than 20 times relative to the original image data in typical reusable-background scenes, before counting the JSCC encoder's own one-third coding-stage compression.
  • Because only the foreground crop enters the deep JSCC encoder, the required GPU memory and inference compute drop, making high-resolution transmission feasible on resource-constrained devices.
  • A base station can choose per user between direct and key-information transmission, trading lower per-user data against image quality, and the paper's dynamic-programming allocation maximizes the sum communication quality under a total power budget.
  • For video with a relatively fixed background, a dynamic background library can be assembled from frames and yields better reconstruction than direct JSCC, with the caveat that performance is below that of a pre-established background library.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper: the claimed 20x compression is essentially the reciprocal of the foreground's fraction of the image; a simple measurable prediction is that the ratio should track the foreground area share, so outdoor scenes with large subjects should show much smaller gains.
  • Beyond the paper: the method implicitly treats the foreground subject as the information that matters; applications where the background carries the semantic content (e.g., environment monitoring, scene analysis) would lose exactly the information the pipeline discards, so the quality metric PSNR may overstate usefulness in those settings.
  • Beyond the paper: the same foreground/background library idea could be applied to any recognizable class of foreground objects (vehicles, products, text) if a segmentation model exists, turning the method into a general semantic-library codec rather than a people-centric one.
  • Beyond the paper: a stress test would compare the pipeline against direct JSCC on images with a deliberately novel background and a small foreground; the fallback to direct transmission means the proposed method should tie, not beat, the baseline there, which would delimit the scope of the claimed advantage.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper proposes a deep joint source-channel coding (JSCC) framework for wireless image transmission that separates foreground and background, transmits only the foreground crop through an adaptive-rate JSCC model, and reconstructs the final image using a pre-shared or dynamically constructed background library. The framework uses Mediapipe for pose detection, Rembg for foreground segmentation, ORB for background matching, and a dynamic programming base-station resource allocation scheme. The authors claim significant PSNR improvements over direct JSCC transmission, especially at low SNR, and a compression factor of more than 20x. The paper also extends the method to video via dynamic background library construction and discusses future generative-AI extensions.

Significance. The problem addressed, efficient transmission of visual information over 6G networks, is relevant and timely. The proposed architecture leverages widely available open-source tools and targets practical scenarios with static or slowly varying backgrounds, such as video streaming and surveillance. If the claimed gains were properly validated, the method could reduce bandwidth and computational requirements. However, the current version does not provide the necessary evidence: the central PSNR claim is based on an unequal comparison, the compression factor is asserted without bit accounting, and the background matching pipeline as written contains an internal inconsistency. No code, numerical tables, or datasets are provided. The strengths are the clearly stated system architecture and the explicit discussion of limitations (e.g., failure fallback in Section II-B and dynamic-background challenges in Section III). Yet, the experimental validation is insufficient for a journal-level claim.

major comments (5)
  1. [§II-B] The masking definitions are internally inconsistent. The paper states that M(x,y)=1 indicates the area to be reserved and M=0 indicates the area to be shielded, with M=0 inside the human hull H. However, the masked image is defined as X_m(x,y)=X(x,y) if M(x,y)=0 and 0 if M(x,y)=1, which keeps the person region and zeros out the background. The same inverted logic is applied to B_m. Consequently, the ORB matching described in this section would compare the person's features against the background's person-area features, contrary to the stated goal of preventing interference from the characters. This error is load-bearing because the entire background-matching procedure and the subsequent experimental interpretation depend on these equations. Please correct the definitions and rerun the matching pipeline.
  2. [§V-A] The PSNR comparison in Figures 6 and 7 is not resource-matched. In the proposed method, the background is taken from a pre-shared library and is never passed through the noisy channel; only the foreground crop is transmitted via JSCC. In the direct JSCC baseline, the full image is transmitted through the same noisy channel. Computing PSNR on the full synthesized image therefore inflates the proposed method's score because the background is noiseless by construction. The paper does not report the total transmitted bits, including background library construction, background index, object coordinates, and crop size, nor does it provide PSNR versus a rate-normalized baseline. Without bit accounting, the claimed "significant improvements" are not established.
  3. [§V-A] The claim that the proposed method "can compress more than 20 times the amount of information compared with the original data transmission" is unsupported. No accounting is given for the background library overhead, the index of the matched background, the coordinates of the bounding box, or the reduction factor of the cropped region. Any end-to-end compression claim requires a complete bit-count comparison, including the cost of establishing and updating the library, which is not provided.
  4. [§II-B and §V-A] The experimental evaluation is subject to selection bias. As stated in Section II-B, when no background satisfies the minimum match threshold N_min, the image is transmitted directly using JSCC. The experiments in Section V-A show only matched-background cases, so the reported PSNR is conditional on matching success. The paper does not report matching success rates or include fallback cases in the aggregate results. Please provide end-to-end average PSNR over an independent test set, explicitly accounting for matched and unmatched images.
  5. [§V-A] The experimental section lacks essential quantitative details. No numeric PSNR values, error bars, or number of test images are given, and the test-set description is missing. Figures 6 and 7 cannot be used to assess statistical significance or reproducibility. A table with mean PSNR and standard deviation for each method and SNR point is necessary to support the central claim.
minor comments (5)
  1. [§II-C] The crop formula X_E = X[x_1:y_1, x_2:y_2] appears to mix coordinate roles; it should likely be X[x_1:x_2, y_1:y_2] given the preceding definitions of (x_1,y_1) and (x_2,y_2) as the upper-left and lower-right corners.
  2. [§IV] The optimization notation is difficult to follow: the summation symbols are typeset incorrectly (e.g., "nP"), and the relationship between the maximum delay T and the dynamic programming recursion is not explained. Clarify the constraints and how T enters the optimization.
  3. [§V] The paper states that the JSCC model is evaluated on CIFAR-10, but the experimental section then refers to "four sets of images from different scenarios" and video frames. Please clarify whether the model was trained and tested on CIFAR-10 or on a different dataset, and describe the test images.
  4. [References] There are citation formatting issues: reference [5] duplicates reference [3], and several reference entries have incomplete author lists or incorrect spacing. Please run a careful reference consistency check.
  5. [General] The paper contains many typographical and punctuation errors (e.g., missing spaces after equations, inconsistent use of commas). A thorough proofreading pass is needed before resubmission.

Circularity Check

1 steps flagged · score 6.0 of 10

Headline PSNR gain is built into the evaluation by reconstructing the background from a noiseless pre-shared library while the baseline transmits the whole image over the noisy channel.

  1. self definitional [Section II-F and Section V-A]
    "After restoring to the corresponding size, we will utilize the background synthesis function in the Rembg library to combine the extracted and transmitted features with the background in the previously matched background library ... we compared the synthesized images after feature extraction and transmission with the images generated directly through the JSCC model. Under different SNR conditions, we compared the Peak Signal-to-Noise Ratio (PSNR) of images produced by the two methods."

    The proposed receiver's output is defined as the transmitted foreground composited onto a library background that never passes through the noisy channel, while the baseline reconstructs the entire image from the same noisy channel. Because PSNR is measured on the final full image, the background portion of the proposed reconstruction is noiseless by construction, whereas the baseline's background is channel-corrupted; at low SNR this protocol difference dominates the reported gap. The paper does not debit the bits for building/updating the library, transmitting the selected-background index, crop coordinates, or resolution metadata, and Section II-B routes non-matching images to direct JSCC, so the evaluation is restricted to matched-background successes.

full rationale

Step 1 is the only construction-level reduction: the headline 'significant improvements in PSNR' is largely determined by the definition of the received image, which includes a noiseless side-information background. The rest of the paper is not circular. The adaptive base-station resource allocation is a standard knapsack dynamic program and does not smuggle in its conclusion. The adaptive-rate JSCC model is adopted from [50] (external), and the self-citations (e.g., [19], [31], [41]) are background/related-work citations, not load-bearing theorems. I also flag two non-circular support problems: the 'more than 20 times' compression claim in Section V-A has no bit budget for library/index/coordinate overhead, and the Section II-B masking equations appear inverted relative to their prose (M=0 is said to be shielded yet X_m retains X where M=0), so the matching pipeline is not well-specified. These affect correctness but are not derivation loops. Score 6 because the central PSNR prediction is forced by the evaluation construction, even though the system concept is not a pure fit.

Assumptions & free parameters 4 free parameters · 6 assumptions · 0 invented entities

The central claim relies on a stack of pre-trained tools and unstated thresholds rather than new mathematical content. There are no invented entities. The free parameters T, N_min, alpha, and the resolution threshold are never given values, which also hurts reproducibility.

free parameters (4)
  • ORB matching threshold T
    Defined in Section II-B to declare a descriptor match; value is not given yet the matching outcome and background selection depend on it.
  • Minimum match threshold N_min
    Defined in Section II-B as the cutoff for accepting a matched background; without a value the fallback behavior is unspecified.
  • Quality factor alpha for key information transmission
    Defined in Section IV as 0<alpha<1; the optimization result depends on this weight but no value or sensitivity analysis is reported.
  • Resolution compression threshold
    Defined in Section II-D as the resolution limit set according to computing power; no value is given.
assumptions (6)
  • domain assumption Mediapipe pose detection reliably finds humans in the transmitted images.
    Used in Section II-B to build the masked image; if detection fails, the foreground crop and background matching are wrong.
  • domain assumption Rembg produces accurate foreground masks.
    Used in Sections II-C and III-B to separate foreground and background; the whole pipeline depends on clean segmentation.
  • domain assumption A matching background exists in the pre-shared library at both ends.
    Section II-A motivates this for static or slowly changing backgrounds; Section II-B states direct transmission when matching fails, so gains are conditional on this assumption.
  • domain assumption ORB descriptors and Hamming distance quantify background similarity well enough for matching.
    Section II-B defines the matching step; no validation of this similarity measure is provided.
  • standard math Shannon capacity formula and knapsack dynamic programming are valid for the power allocation subproblem.
    Section IV derives optimal transmission decisions from r_n = log2(1 + P_n G L / N_n); this is standard, but the model ignores rate-distortion tradeoffs of the two modes.
  • domain assumption The deep JSCC model from [50] trained on CIFAR-10 generalizes to arbitrary foreground crops.
    Section V describes training on CIFAR-10 and then applying the model to unspecified test images; no adaptation for crop size or domain shift is described.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Semantic-Aware Visual Information Transmission With Key Information Extraction Over Wireless Networks." pith.science (2026). https://pith.science/paper/YRJGQAM7

@misc{pith2026250612786,
  author       = {Pith},
  title        = {Pith review of: Semantic-Aware Visual Information Transmission With Key Information Extraction Over Wireless Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YRJGQAM7}},
  note         = {Machine review of arXiv:2506.12786}
}
read the original abstract

The advent of 6G networks demands unprecedented levels of intelligence, adaptability, and efficiency to address challenges such as ultra-high-speed data transmission, ultra-low latency, and massive connectivity in dynamic environments. Traditional wireless image transmission frameworks, reliant on static configurations and isolated source-channel coding, struggle to balance computational efficiency, robustness, and quality under fluctuating channel conditions. To bridge this gap, this paper proposes an AI-native deep joint source-channel coding (JSCC) framework tailored for resource-constrained 6G networks. Our approach integrates key information extraction and adaptive background synthesis to enable intelligent, semantic-aware transmission. Leveraging AI-driven tools, Mediapipe for human pose detection and Rembg for background removal, the model dynamically isolates foreground features and matches backgrounds from a pre-trained library, reducing data payloads while preserving visual fidelity. Experimental results demonstrate significant improvements in peak signal-to-noise ratio (PSNR) compared with traditional JSCC method, especially under low-SNR conditions. This approach offers a practical solution for multimedia services in resource-constrained mobile communications.

Figures

Figures reproduced from arXiv: 2506.12786 by the authors.

Figure 1
Figure 1. Structure of Our Key Information Extraction and Background Synthesis Model. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Structure of the considered JSCC model. The structure of the considered model is shown in [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Get the maximum background information by integrating the back [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Result of image with JSCC method and PM. [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Result of image with JSCC method and PM. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Performance of different images with JSCC method. [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Performance of different images with proposed method. [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

55 extracted references · 42 canonical work pages

  1. [1]

    What should 6g be?

    S. Dang, O. Amin, B. Shihada, and et al., “What should 6g be?”Nature Electronics, vol. 3, no. 1, pp. 20–29, 2020

    work page 2020

  2. [2]

    The road towards 6g: A comprehensive survey,

    W. Jiang, B. Han, M. A. Habibi, and et al., “The road towards 6g: A comprehensive survey,”IEEE Open Journal of the Communications Society, vol. 2, pp. 334–366, 2021

    work page 2021

  3. [4]

    A joint communication and computation design for distributed RISs assisted probabilistic semantic communication in IIoT,

    Z. Zhao, Z. Yang, C. Huang, L. Wei, Q. Yang, C. Zhong, W. Xu, and Z. Zhang, “A joint communication and computation design for distributed RISs assisted probabilistic semantic communication in IIoT,” IEEE Internet Things J., vol. 11, no. 16, pp. 26 568–26 579, Aug. 2024

    work page 2024

  4. [5]

    A vision of 6g wireless systems: Applications, trends, technologies, and open research problems,

    W. Saad, M. Bennis, and M. Chen, “A vision of 6g wireless systems: Applications, trends, technologies, and open research problems,”IEEE network, vol. 34, no. 3, pp. 134–142, 2019

    2019

  5. [6]

    A survey on 5g: The next generation of mobile communication,

    N. Panwar, S. Sharma, and A. K. Singh, “A survey on 5g: The next generation of mobile communication,”Physical Communication, vol. 18, pp. 64–84, 2016

    work page 2016

  6. [7]

    6g internet of things: A comprehensive survey,

    D. C. Nguyen, M. Ding, P. N. Pathirana, A. Seneviratne, J. Li, D. Niyato, O. Dobre, and H. V . Poor, “6g internet of things: A comprehensive survey,”IEEE Internet of Things Journal, vol. 9, no. 1, pp. 359–383, 2021

    2021

  7. [8]

    A joint communication and computation design for probabilistic semantic com- munications,

    Z. Zhao, Z. Yang, M. Chen, Z. Zhang, and H. V . Poor, “A joint communication and computation design for probabilistic semantic com- munications,”Entropy, vol. 26, no. 5, Apr. 2024

    work page 2024

  8. [9]

    Communications in the 6g era,

    H. Viswanathan and P. E. Mogensen, “Communications in the 6g era,” IEEE access, vol. 8, pp. 57 063–57 074, 2020

    work page 2020

Show all 55 references
  1. [10]

    6g wireless systems: Vi- sion, requirements, challenges, insights, and opportunities,

    H. Tataria, M. Shafi, A. F. Molisch, and et al., “6g wireless systems: Vi- sion, requirements, challenges, insights, and opportunities,”Proceedings of the IEEE, vol. 109, no. 7, pp. 1166–1199, 2021

    2021

  2. [11]

    A joint communication and computation design for semantic wireless communication with probability graph,

    Z. Zhao, Z. Yang, X. Gan, Q.-V . Pham, C. Huang, W. Xu, and Z. Zhang, “A joint communication and computation design for semantic wireless communication with probability graph,”J. Franklin Inst., vol. 361, no. 13, p. 107055, Sep. 2024

    2024

  3. [12]

    A mathematical theory of communication,

    C. E. Shannon, “A mathematical theory of communication,”The Bell system technical journal, vol. 27, no. 3, pp. 379–423, 1948

    1948

  4. [13]

    Computer vision for assistive technologies,

    M. Leo, G. Medioni, M. Trivedi, T. Kanade, and G. M. Farinella, “Computer vision for assistive technologies,”Computer Vision and Image Understanding, vol. 154, pp. 1–15, 2017

    2017

  5. [14]

    Semantic communications for future internet: Fundamentals, applications, and challenges,

    W. Yang, H. Du, Z. Q. Liew, W. Y . B. Lim, Z. Xiong, D. Niyato, X. Chi, X. Shen, and C. Miao, “Semantic communications for future internet: Fundamentals, applications, and challenges,”IEEE Communications Surveys & Tutorials, vol. 25, no. 1, pp. 213–250, 2022

    2022

  6. [15]

    Applying deep-learning-based com- puter vision to wireless communications: Methodologies, opportunities, and challenges,

    Y . Tian, G. Pan, and M. S. Alouini, “Applying deep-learning-based com- puter vision to wireless communications: Methodologies, opportunities, and challenges,”IEEE Open Journal of the Communications Society, vol. 2, pp. 132–143, 2020

    2020

  7. [16]

    Semantic communications: Overview, open issues, and future research directions,

    X. Luo, H. H. Chen, and Q. Guo, “Semantic communications: Overview, open issues, and future research directions,”IEEE Wireless Communi- cations, vol. 29, no. 1, pp. 210–219, 2022

    2022

  8. [17]

    Towards a theory of semantic communication,

    J. Bao, P. Basu, M. Dean, C. Partridge, A. Swami, W. Leland, and J. A. Hendler, “Towards a theory of semantic communication,” in2011 IEEE Network Science Workshop, 2011, pp. 110–117

    2011

  9. [18]

    Preserving quality of information by using semantic relationships,

    P. Basu, J. Bao, M. Dean, and J. Hendler, “Preserving quality of information by using semantic relationships,”Pervasive and Mobile Computing, vol. 11, pp. 188–202, 2014. 10

    2014

  10. [19]

    Compression ratio allocation for probabilistic semantic communication with RSMA,

    Z. Zhao, Z. Yang, Y . Hu, C. Zhu, M. Shikh-Bahaei, W. Xu, Z. Zhang, and K. Huang, “Compression ratio allocation for probabilistic semantic communication with RSMA,”IEEE Trans. Commun., pp. 1–1, 2025

    2025

  11. [20]

    Transmit what you need: task-adaptive semantic communications for visual information,

    J. Park and S. W. Yoon, “Transmit what you need: task-adaptive semantic communications for visual information,”arXiv preprint arXiv:2412.13646, 2024

    2024

  12. [21]

    Unpaired image-to-image translation using cycle-consistent adversarial networks,

    J. Y . Zhu, T. Park, P. Isola, and et al., “Unpaired image-to-image translation using cycle-consistent adversarial networks,” inProceedings of the IEEE international conference on computer vision, 2017, pp. 2223–2232

    2017

  13. [22]

    Generative ai meets semantic communication: Evolution and revolution of communication tasks,

    E. Grassucci, J. Park, S. Barbarossa, and et al., “Generative ai meets semantic communication: Evolution and revolution of communication tasks,”arXiv preprint arXiv:2401.06803, 2024

    2024 arXiv

  14. [23]

    Generative ai for semantic communication: Architecture, challenges, and outlook,

    L. Xia, Y . Sun, C. Liang, and et al., “Generative ai for semantic communication: Architecture, challenges, and outlook,”IEEE Wireless Communications, vol. 32, no. 1, pp. 132–140, 2025

    2025

  15. [24]

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

    C. Liang, H. Du, Y . Sun, and et al., “Generative ai-driven semantic communication networks: Architecture, technologies and applications,” IEEE Transactions on Cognitive Communications and Networking, 2024

    2024

  16. [25]

    Joint source-channel coding and unequal error protection for video plus depth,

    A. V osoughi, P. C. Cosman, and L. B. Milstein, “Joint source-channel coding and unequal error protection for video plus depth,”IEEE Signal Processing Letters, vol. 22, no. 1, pp. 31–34, 2014

    2014

  17. [26]

    Transactions papers source-optimized channel coding for digital transmission channels,

    S. Heinen and P. Vary, “Transactions papers source-optimized channel coding for digital transmission channels,”IEEE transactions on commu- nications, vol. 53, no. 4, pp. 592–600, 2005

    2005

  18. [27]

    Combined vector quantisation and index assignment with embedded redundancy for noisy channels,

    V . Bozantizis and F. Ali, “Combined vector quantisation and index assignment with embedded redundancy for noisy channels,”Electronics Letters, vol. 36, no. 20, pp. 1711–1713, 2000

    2000

  19. [28]

    Joint source-channel vector quantization for compressed sensing,

    A. Shirazinia, S. Chatterjee, and M. Skoglund, “Joint source-channel vector quantization for compressed sensing,”IEEE transactions on signal processing, vol. 62, no. 14, pp. 3667–3681, 2014

    2014

  20. [29]

    A linear encoding approach to index assignment in lossy source-channel coding,

    M. Fresia and G. Caire, “A linear encoding approach to index assignment in lossy source-channel coding,”IEEE transactions on information theory, vol. 56, no. 3, pp. 1322–1344, 2010

    2010

  21. [30]

    Evaluation of analog joint source-channel coding systems for multiple access channels,

    O. Fresnedo, J. P. Gonzalez-Coma, M. Hassanin, and et al., “Evaluation of analog joint source-channel coding systems for multiple access channels,”IEEE Transactions on Communications, vol. 63, no. 6, pp. 2312–2324, 2015

    2015

  22. [31]

    Sum rate maximization for distributed riss assisted probabilistic se- mantic communication,

    Z. Zhao, Z. Yang, Q. Yang, C. Huang, M. Shikh-Bahaei, and Z. Zhang, “Sum rate maximization for distributed riss assisted probabilistic se- mantic communication,” in2024 IEEE 34th International Workshop on Machine Learning for Signal Processing (MLSP), 2024, pp. 1–6

    2024

  23. [32]

    Deep learning based communication over the air,

    S. D ¨orner, S. Cammerer, J. Hoydis, and et al., “Deep learning based communication over the air,”IEEE Journal of Selected Topics in Signal Processing, vol. 12, no. 1, pp. 132–143, 2017

    2017

  24. [33]

    Graph-based deep learning for communication networks: A survey,

    W. Jiang, “Graph-based deep learning for communication networks: A survey,”Computer Communications, vol. 185, pp. 40–54, 2022

    2022

  25. [34]

    Communication algorithms via deep learning,

    H. Kim, Y . Jiang, R. Rana, and et al., “Communication algorithms via deep learning,”arXiv preprint arXiv:1805.09317, 2018

    2018 arXiv

  26. [35]

    Deep learning for wireless com- munications: An emerging interdisciplinary paradigm,

    L. Dai, R. Jiao, F. Adachi, and et al., “Deep learning for wireless com- munications: An emerging interdisciplinary paradigm,”IEEE Wireless Communications, vol. 27, no. 4, pp. 133–139, 2020

    2020

  27. [36]

    Deep learning enabled seman- tic communication systems,

    H. Xie, Z. Qin, G. Y . Li, and et al., “Deep learning enabled seman- tic communication systems,”IEEE transactions on signal processing, vol. 69, pp. 2663–2675, 2021

    2021

  28. [37]

    Deep learning-based end-to- end wireless communication systems with conditional gans as unknown channels,

    H. Ye, L. Liang, G. Y . Li, and et al., “Deep learning-based end-to- end wireless communication systems with conditional gans as unknown channels,”IEEE Transactions on Wireless Communications, vol. 19, no. 5, pp. 3133–3143, 2020

    2020

  29. [38]

    Deep learning for wireless physical layer: Opportunities and challenges,

    T. Wang, C. K. Wen, H. Wang, and et al., “Deep learning for wireless physical layer: Opportunities and challenges,”China Communications, vol. 14, no. 11, pp. 92–111, 2017

    2017

  30. [39]

    Deep learning for physical-layer 5g wireless techniques: Opportunities, challenges and solutions,

    H. Huang, S. Guo, G. Gui, and et al., “Deep learning for physical-layer 5g wireless techniques: Opportunities, challenges and solutions,”IEEE Wireless Communications, vol. 27, no. 1, pp. 214–222, 2019

    2019

  31. [40]

    Communication-efficient distributed deep learning: A comprehensive survey,

    Z. Tang, S. Shi, W. Wang, and et al., “Communication-efficient distributed deep learning: A comprehensive survey,”arXiv preprint arXiv:2003.06307, 2020

    2003 arXiv

  32. [41]

    Energy-efficient probabilistic semantic communication over space-air- ground integrated networks,

    Z. Zhao, Z. Yang, M. Chen, C. Zhu, W. Xu, Z. Zhang, and K. Huang, “Energy-efficient probabilistic semantic communication over space-air- ground integrated networks,”IEEE Trans. Wireless Commun., pp. 1–1, 2025

    2025

  33. [42]

    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 Transactions on Cognitive Communications and Networking, vol. 5, no. 3, pp. 567–579, 2019

    2019

  34. [43]

    Successive refinement of images with deep joint source-channel coding,

    D. B. Kurka and D. G ¨und¨uz, “Successive refinement of images with deep joint source-channel coding,” in2019 IEEE 20th International Workshop on Signal Processing Advances in Wireless Communications (SPA WC), 2019, pp. 1–5

    2019

  35. [44]

    Deepjscc-f: Deep joint source-channel coding of images with feedback,

    ——, “Deepjscc-f: Deep joint source-channel coding of images with feedback,”IEEE Journal on Selected Areas in Information Theory, 2020

    2020

  36. [45]

    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 Transactions on Circuits and Systems for Video Tech- nology, 2021

    2021

  37. [46]

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

    M. Song, N. Ma, C. Dong, and et al., “Deep joint source-channel coding for wireless image transmission with adaptive models,”Electronics, vol. 12, no. 22, p. 4637, 2023

    2023

  38. [47]

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

    W. Chen, Y . Chen, Q. Yang, and et al., “Deep joint source-channel coding for wireless image transmission with entropy-aware adaptive rate control,” inGLOBECOM 2023-2023 IEEE Global Communications Conference. IEEE, 2023, pp. 2239–2244

    2023

  39. [48]

    Joint source- channel coding for wireless image transmission: A deep compressed- sensing based method,

    M. A. Jarrahi, E. Bourtsoulatze, and V . Abolghasemi, “Joint source- channel coding for wireless image transmission: A deep compressed- sensing based method,” in2024 IEEE Wireless Communications and Networking Conference (WCNC). IEEE, 2024, pp. 1–6

    2024

  40. [49]

    Neural joint sourcechannel coding,

    K. Choi, K. Tatwawadi, A. Grover, T. Weissman, and S. Ermon, “Neural joint sourcechannel coding,” inInternational Conference on Machine Learning. PMLR, 2019, pp. 1182–1192

    2019

  41. [50]

    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,” inICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Pro- cessing (ICASSP). IEEE, 2022, pp. 5193–5197

    2022

  42. [51]

    Deep joint source-channel coding using overlap image division for block noise reduction,

    R. Yamamoto, Y . Inoue, and D. Hisano, “Deep joint source-channel coding using overlap image division for block noise reduction,” in2024 IEEE 99th V ehicular Technology Conference (VTC2024-Spring). IEEE, 2024, pp. 1–6

    2024

  43. [52]

    Deep learning-based im- age compression for wireless communications: Impacts on reliability, throughput, and latency,

    M. Naseri, P. Ashtari, M. Seif, and et al., “Deep learning-based im- age compression for wireless communications: Impacts on reliability, throughput, and latency,”arXiv preprint arXiv:2411.10650, 2024

    2024 arXiv

  44. [53]

    Mediapipe: A framework for building perception pipelines,

    C. Lugaresi, J. Tang, H. Nash, and et al., “Mediapipe: A framework for building perception pipelines,”arXiv preprint arXiv:1906.08172, 2019

    1906 arXiv

  45. [54]

    Mediapipe: A framework for perceiving and processing reality,

    ——, “Mediapipe: A framework for perceiving and processing reality,” inThird workshop on computer vision for AR/VR at IEEE computer vision and pattern recognition (CVPR), 2019

    2019

  46. [55]

    Orb: An efficient alternative to sift or surf,

    E. Rublee, V . Rabaud, K. Konolige, and et al., “Orb: An efficient alternative to sift or surf,” in2011 International conference on computer vision. Ieee, 2011, pp. 2564–2571

    2011

  47. [56]

    Real-time intermediate flow estimation for video frame interpolation,

    Z. Huang, T. Zhang, W. Heng, and et al., “Real-time intermediate flow estimation for video frame interpolation,” inEuropean Conference on Computer Vision. Cham: Springer Nature Switzerland, 2022, pp. 624– 642

    2022

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.