Pith. sign in

REVIEW 4 major objections 6 minor 32 references

Two-flow Feedback Multi-scale Progressive Generative Adversarial Network

T0 review · 4 major / 6 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read This paper claims a two-flow feedback GAN architecture, MSPG-SEN, improves image quality, stability, and cost, and reports state-of-the-art scores on five datasets.

desk verdict The paper's central claim rests on numbers that aren't in the manuscript: Section V is empty, the abstract names placeholder datasets, and no evaluation exists. read the letter →

arxiv 2508.16089 v1 pith:4JJAU66L submitted 2025-08-22 cs.CV cs.AI

Sun Weikai , Song Shijie , Chi Wenjie This is my paper
classification cs.CVcs.AI
keywords convergenceandstabilityrobustregressionimageprocessingcomputervisionneuralnetsmultimodaldeeplearninggenerativeadversarialnetworksgenerationattentionmechanism
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 proposes MSPG-SEN, a generative adversarial network that combines two parallel processing flows, a multi-scale progressive generator, a dynamic attention mechanism, a two-flow residual network, and a perception-behavior feedback loop with a DQN-based balancer. The authors claim this architecture improves image quality and training stability while reducing training cost, and they report state-of-the-art percentages on five datasets named in the abstract. However, the manuscript contains no experimental results section: Section V is empty, and the datasets named in the abstract (INKK, AWUN, IONJ, POKL, OPIN) do not match the six datasets listed in Section IV (coco2017, CUB 200-2011, vangogh2photo, summer2winter yosemite, grumpifycat, monet2photo). A sympathetic reader would take the architectural proposal as the contribution, but the empirical claims are the load-bearing part and currently lack documented support.

What carries the argument

MSPG-SEN is the full architecture. Its carriers are: (1) DEMA, an attention mechanism that dynamically fuses channel and spatial attention with implicit context embedding and a contrastive loss to separate focused and expanded features; (2) GCTDRN, a two-flow residual block that fuses branches of kernel sizes 3×3, 5×5, and 7×7 with a shortcut connection; (3) APFL, a meta-learning feedback loop that adjusts learning rates and losses based on performance indicators; and (4) a DQN balancer that treats GAN training as a reinforcement-learning problem to keep generator and discriminator in balance.

What would settle it

Open Section V (Experimental Results): it is empty. A re-run on the five datasets named in the abstract with the stated metrics would either reproduce the reported percentages or not; until that section contains a table with numbers, the claimed state-of-the-art results have no observable support.

Watch

Extended reading notes

Core claim

The paper argues that a GAN whose generator runs two parallel multi-scale residual flows, whose attention is a dynamically fused channel/spatial mechanism with a contrastive separation loss, and whose generator-discriminator interplay is regulated by a perception-behavior feedback loop plus a DQN balancer, will generate images of higher quality and diversity while training more stably and cheaply than existing GANs. It calls this architecture MSPG-SEN and reports state-of-the-art scores on five datasets named in the abstract; those scores are not shown in the manuscript.

Load-bearing premise

The paper's central claim depends on five dataset scores (89.7%, 78.3%, 85.5%, 88.7%, 96.4%) being real measurements from a defined evaluation protocol; the current manuscript provides no such protocol or results.

Editorial extensions

If this is right

  • If MSPG-SEN works as described, GAN training no longer needs hand-tuned balancing schedules; APFL and the DQN balancer would automate the generator-discriminator trade-off.
  • The DEMA attention module could be extracted and reused in other generators or image-restoration networks, since its design is task-agnostic.
  • The two-flow residual fusion would give generators a concrete way to combine local and global features at multiple scales, which is directly relevant to high-resolution synthesis.
  • A stable training wrapper would lower the computing cost of producing high-quality images, making GANs more accessible when diffusion models are too expensive.
  • The adversarial feature-enhancement module, an auxiliary discriminator inside the generator, offers a mechanism specifically aimed at suppressing mode collapse while preserving diversity.

Reading between the lines

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

  • I infer that the mismatch between the five datasets named in the abstract and the six datasets named in Section IV means the reported percentages cannot be traced to a specific evaluation protocol; the reader should treat them as unverified.
  • A natural test of the architectural claim is to ablate each module separately on a standard benchmark (e.g., CIFAR-10 or ImageNet) with FID and recall; the paper claims ablations were done but does not report them.
  • If the DQN balancer is genuinely effective, it could be applied as a wrapper to existing GAN architectures without changing their generators, which would be a cheap way to test the claim independently.
  • I also infer that the phrase 'only 88.7% with INJK' in the abstract is likely a leftover placeholder, which reinforces the need for a clean, complete experimental write-up before the central claim can be assessed.
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

4 major / 6 minor

Summary. The paper proposes MSPG-SEN, a generative adversarial network architecture combining a two-flow feedback multi-scale progressive design, a DEMA attention mechanism, a GCTDRN residual network, an APFL feedback loop, and a DQN-based BALANCE balancer. The abstract and introduction claim state-of-the-art generation results on five datasets with reported percentages (INKK 89.7%, AWUN 78.3%, IONJ 85.5%, POKL 88.7%, OPIN 96.4%). Section IV describes experimental settings on six public datasets (coco2017, CUB 200-2011, vangogh2photo, summer2winter yosemite, grumpifycat, monet2photo) that are scrambled, mixed, and cleaned. However, Section V, titled 'EXPERIMENTAL RESULTS', is empty. No quantitative or qualitative results, baselines, evaluation metrics, error bars, or protocol are provided anywhere in the manuscript. The central claim is therefore unsupported.

Significance. If the claimed results were real and reproducible, the proposed mechanisms—especially the DQN balancer and the APFL feedback loop—could be of interest to the GAN community. However, the manuscript provides no experimental evidence, no code, and no parameter-free derivations. The only verification available would be the missing experimental section. As submitted, the contribution cannot be assessed, and the significance of the work is effectively nil because the central claim is not backed by any data.

major comments (4)
  1. [V. EXPERIMENTAL RESULTS] Section V is empty. The central claim of state-of-the-art performance on five datasets is therefore entirely unsupported. The percentages in the abstract (INKK 89.7%, AWUN 78.3%, IONJ 85.5%, POKL 88.7%, OPIN 96.4%) are not associated with any metric, dataset, or evaluation protocol, making them impossible to verify or interpret.
  2. [IV. EXPERIMENTAL SETTINGS] The abstract names five datasets (INKK, AWUN, IONJ, POKL, OPIN), but Section IV lists six different public datasets (coco2017, CUB 200-2011, vangogh2photo, summer2winter yosemite, grumpifycat, monet2photo) and states that they were 'scrambled and mixed' and cleaned. No mapping between these six datasets and the five claimed benchmark datasets is given. This inconsistency makes the claimed state-of-the-art results unverifiable and suggests the percentages are placeholders.
  3. [IV. EXPERIMENTAL SETTINGS / V. EXPERIMENTAL RESULTS] No evaluation protocol is defined. The text says 'For quantitative and qualitative comparisons' but no baseline methods, no metrics (e.g., FID, IS, accuracy), no data splits, and no error bars are presented. Even if Section V contained numbers, the absence of a defined metric and baselines would make them meaningless as evidence of state-of-the-art performance.
  4. [III-B, Eq. (18)] The auxiliary discriminator loss in Eq. (18) is contradictory: LDaux = -E[log Daux(Fgen)] - E[log(1 - Daux(Fgen))]. Both terms are evaluated on the same generated feature Fgen, and they respectively encourage Daux to classify Fgen as real and as fake. This loss cannot be optimized as written and undermines the claimed benefit of the AFE module for diversity and mode-collapse prevention.
minor comments (6)
  1. [III-A, Eq. (7)] The contrast loss denominator is malformed: it appears as 'sum_j exp(sim(Fi, Pj)/tau)' but the numerator uses Fi and Pi; the notation should clarify that the sum over j runs over both positive and negative samples, and parentheses are missing.
  2. [General] Equation numbering is out of order: Eq. (8)–(14) appear after Eq. (15)–(20). This makes the paper difficult to follow.
  3. [II-A] Typo: 'BANLANCE' should be 'BALANCE'.
  4. [References] The reference heading 'REFERENCES' appears twice.
  5. [III-C] Figure references are inconsistent: the text says 'Figure 2 shows the APFL feedback loop' and 'Figure 3 shows the meta-learning module,' but the captions indicate Figure 3 is the APFL framework and Figure 4 is the meta-learning architecture.
  6. [Abstract] The abstract contains an incomplete phrase 'only 88.7% with INJK' with no dataset name or context.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular reasoning identified; the paper's failure is missing evidence, not a derivation that reduces to its own inputs.

full rationale

The derivation chain in this manuscript consists of standard GAN losses (Eqs. 10-11, 17-20), a feature-matching loss (Eq. 12), a contrastive loss (Eq. 7), and compositional blocks (Eqs. 1-6, 15-16). None of these equations is defined in terms of the claimed state-of-the-art percentages, and neither the architecture components nor the losses are fitted to those percentages. The abstract's dataset names and scores (INKK 89.7%, AWUN 78.3%, IONJ 85.5%, POKL 88.7%, OPIN 96.4%) do not appear in the experimental settings; Section IV names six different public datasets and Section V, headed 'EXPERIMENTAL RESULTS', is empty. This is a severe absence-of-evidence and verifiability problem, but it is not circularity: the central claim is unsupported rather than self-supporting. There are also no self-citations that carry a load-bearing argument; all cited references are external prior work, and no uniqueness theorem or ansatz is imported from the authors' own prior publications. Accordingly, no step in the paper reduces by construction to its own inputs, so the circularity score is 0.

Assumptions & free parameters 5 free parameters · 6 assumptions · 5 invented entities

All four claimed contributions are modules proposed in Section III, and none of them is evaluated in the paper. The free parameters are standard hyperparameters and learnable weights with no tuning protocol. The invented entities have no falsifiable handle outside the manuscript because no code, data, or experimental results are provided.

free parameters (5)
  • lambda_aux
    Hyperparameter balancing generator loss and auxiliary discriminator loss in Eq. 20; no value or tuning procedure given.
  • alpha, beta
    Learnable fusion weights in DEMA (Eq. 6) and GCTDRN (Eq. 15); no initialization or final values reported.
  • tau
    Temperature parameter in the contrastive loss (Eq. 7); not specified.
  • learning rate eta and decay gamma = initial lr 0.1, EMA 0.9999
    Reported in Section IV but the adjustment schedule is left to the balancer and is not reproducible.
  • drop probability = 0.1
    Set to 10 percent in Section IV without justification.
assumptions (6)
  • domain assumption Adversarial training with the listed generator and discriminator losses converges to a useful equilibrium.
    Invoked in Section III-C and III-B via Eqs. 10-11 and 17-20; no proof or stability analysis is provided.
  • domain assumption Meta-learning with feature matching (Eq. 12) improves GAN training stability.
    Asserted in Section III-C; no theorem, ablation, or external evidence is supplied.
  • domain assumption A DQN-based balancer can improve generator-discriminator balance.
    Proposed in Section III-D; the mechanism is described but never evaluated.
  • ad hoc to paper Scrambling and mixing six public datasets yields a valid benchmark for the five claimed datasets.
    Section IV describes mixing coco2017, CUB, vangogh2photo, etc., but the abstract reports results on INKK, AWUN, IONJ, POKL, and OPIN with no mapping.
  • domain assumption Contrastive loss (Eq. 7) encourages complementary global-local feature learning.
    Stated in Section III-A without experimental verification.
  • domain assumption Multi-branch residual convolution with shortcuts prevents mode collapse and network degradation.
    Used in Eq. 16 and discussed in Section III-B; the claim is standard but not demonstrated here.
invented entities (5)
  • DEMA attention mechanism
    purpose: Capture global-local information and improve feature separation.
    Introduced in Section III-A with no external validation, code, or ablation results.
  • GCTDRN residual network
    purpose: Extract and fuse multi-scale features with global linkage.
    Described in Section III-B; no experimental evidence that it improves training efficiency or generalization.
  • APFL feedback loop
    purpose: Dynamically adjust generator and discriminator training via meta-learning.
    Proposed in Section III-C; no measurements of robustness or training cost reduction.
  • BALANCE (DQN balancer)
    purpose: Use deep Q-networks as a referee to balance generator and discriminator.
    Presented in Section III-D; no ablation or quantitative result supports the claimed stability improvement.
  • AFE (adversarial feature enhancement) module
    purpose: Reduce mode collapse through an auxiliary discriminator on generator features.
    Introduced in Section III-B with no experiment or comparison.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Two-flow Feedback Multi-scale Progressive Generative Adversarial Network." pith.science (2026). https://pith.science/paper/4JJAU66L

@misc{pith2026250816089,
  author       = {Pith},
  title        = {Pith review of: Two-flow Feedback Multi-scale Progressive Generative Adversarial Network},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4JJAU66L}},
  note         = {Machine review of arXiv:2508.16089}
}
read the original abstract

Although diffusion model has made good progress in the field of image generation, GAN\cite{huang2023adaptive} still has a large development space due to its unique advantages, such as WGAN\cite{liu2021comparing}, SSGAN\cite{guibas2021adaptive} \cite{zhang2022vsa} \cite{zhou2024adapt} and so on. In this paper, we propose a novel two-flow feedback multi-scale progressive generative adversarial network (MSPG-SEN) for GAN models. This paper has four contributions: 1) : We propose a two-flow feedback multi-scale progressive Generative Adversarial network (MSPG-SEN), which not only improves image quality and human visual perception on the basis of retaining the advantages of the existing GAN model, but also simplifies the training process and reduces the training cost of GAN networks. Our experimental results show that, MSPG-SEN has achieved state-of-the-art generation results on the following five datasets,INKK The dataset is 89.7\%,AWUN The dataset is 78.3\%,IONJ The dataset is 85.5\%,POKL The dataset is 88.7\%,OPIN The dataset is 96.4\%. 2) : We propose an adaptive perception-behavioral feedback loop (APFL), which effectively improves the robustness and training stability of the model and reduces the training cost. 3) : We propose a globally connected two-flow dynamic residual network(). After ablation experiments, it can effectively improve the training efficiency and greatly improve the generalization ability, with stronger flexibility. 4) : We propose a new dynamic embedded attention mechanism (DEMA). After experiments, the attention can be extended to a variety of image processing tasks, which can effectively capture global-local information, improve feature separation capability and feature expression capabilities, and requires minimal computing resources only 88.7\% with INJK With strong cross-task capability.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

32 extracted references · 23 canonical work pages

  1. [1]

    Vlmo: Unified vision-language pre-training with mixture-of-modality- experts

    Hangbo Bao, Wenhui Wang, Li Dong, Qiang Liu, Owais Khan Mo- hammed, Kriti Aggarwal, Subhojit Som, Songhao Piao, and Furu Wei. Vlmo: Unified vision-language pre-training with mixture-of-modality- experts. Advances in Neural Information Processing Systems, 35:32897– 32912, 2022

    work page 2022

  2. [2]

    Automatic image processing algorithm for light en- vironment optimization based on multimodal neural network model

    Mujun Chen. Automatic image processing algorithm for light en- vironment optimization based on multimodal neural network model. Computational Intelligence and Neuroscience , 2022(1):5156532, 2022

    work page 2022

  3. [3]

    Adaptive fourier neural operators: Efficient token mixers for transformers

    John Guibas, Morteza Mardani, Zongyi Li, Andrew Tao, Anima Anand- kumar, and Bryan Catanzaro. Adaptive fourier neural operators: Efficient token mixers for transformers. arXiv preprint arXiv:2111.13587 , 2021

    arXiv 2021

  4. [4]

    Crd-cgan: Category- consistent and relativistic constraints for diverse text-to-image genera- tion

    Tao Hu, Chengjiang Long, and Chunxia Xiao. Crd-cgan: Category- consistent and relativistic constraints for diverse text-to-image genera- tion. Frontiers of Computer Science , 18(1):181304, 2024. TWO-FLOW FEEDBACK MULTI-SCALE PROGRESSIVE GENERATIVE ADVERSARIAL NETWORK 8

    work page 2024

  5. [5]

    Language is not all you need: Aligning perception with language models

    Shaohan Huang, Li Dong, Wenhui Wang, Yaru Hao, Saksham Singhal, Shuming Ma, Tengchao Lv, Lei Cui, Owais Khan Mohammed, Barun Patra, et al. Language is not all you need: Aligning perception with language models. Advances in Neural Information Processing Systems , 36:72096–72109, 2023

    work page 2023

  6. [6]

    Adaptive frequency filters as efficient global token mixers

    Zhipeng Huang, Zhizheng Zhang, Cuiling Lan, Zheng-Jun Zha, Yan Lu, and Baining Guo. Adaptive frequency filters as efficient global token mixers. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 6049–6059, 2023

    work page 2023

  7. [7]

    Fast and efficient image generation using variational autoencoders and k-nearest neighbor oversampling approach

    Ashhadul Islam and Samir Brahim Belhaouari. Fast and efficient image generation using variational autoencoders and k-nearest neighbor oversampling approach. IEEE Access, 11:28416–28426, 2023

    work page 2023

  8. [8]

    An underwater image enhancement method for a preprocessing framework based on generative adversarial network

    Xiao Jiang, Haibin Yu, Yaxin Zhang, Mian Pan, Zhu Li, Jingbiao Liu, and Shuaishuai Lv. An underwater image enhancement method for a preprocessing framework based on generative adversarial network. Sensors, 23(13):5774, 2023

    work page 2023

Show all 32 references
  1. [9]

    Vilt: Vision-and-language transformer without convolution or region supervision

    Wonjae Kim, Bokyung Son, and Ildoo Kim. Vilt: Vision-and-language transformer without convolution or region supervision. In International conference on machine learning , pages 5583–5594. PMLR, 2021

    2021

  2. [10]

    Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models

    Junnan Li, Dongxu Li, Silvio Savarese, and Steven Hoi. Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models. In International conference on machine learning, pages 19730–19742. PMLR, 2023

    2023

  3. [11]

    Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation

    Junnan Li, Dongxu Li, Caiming Xiong, and Steven Hoi. Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation. In International conference on machine learning, pages 12888–12900. PMLR, 2022

    2022

  4. [12]

    Align before fuse: Vision and language representation learning with momentum distillation

    Junnan Li, Ramprasaath Selvaraju, Akhilesh Gotmare, Shafiq Joty, Caiming Xiong, and Steven Chu Hong Hoi. Align before fuse: Vision and language representation learning with momentum distillation. Ad- vances in neural information processing systems , 34:9694–9705, 2021

    2021

  5. [13]

    Vision- language foundation models as effective robot imitators

    Xinghang Li, Minghuan Liu, Hanbo Zhang, Cunjun Yu, Jie Xu, Hongtao Wu, Chilam Cheang, Ya Jing, Weinan Zhang, Huaping Liu, et al. Vision- language foundation models as effective robot imitators. arXiv preprint arXiv:2311.01378, 2023

    2023 arXiv

  6. [14]

    Scaling language-image pre-training via masking

    Yanghao Li, Haoqi Fan, Ronghang Hu, Christoph Feichtenhofer, and Kaiming He. Scaling language-image pre-training via masking. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 23390–23400, 2023

    2023

  7. [15]

    Novel creation method of feature graphics for image generation based on deep learning algorithms

    Ying Li and Ye Tang. Novel creation method of feature graphics for image generation based on deep learning algorithms. Mathematics, 11(7):1644, 2023

    2023

  8. [16]

    Comparing recognition performance and robustness of multimodal deep learning models for multimodal emotion recognition

    Wei Liu, Jie-Lin Qiu, Wei-Long Zheng, and Bao-Liang Lu. Comparing recognition performance and robustness of multimodal deep learning models for multimodal emotion recognition. IEEE Transactions on Cognitive and Developmental Systems , 14(2):715–729, 2021

    2021

  9. [17]

    Cogan: Cooperatively trained conditional and unconditional gan for person image generation

    Yang Liu, Hao Sheng, Shuai Wang, Yubin Wu, and Zhang Xiong. Cogan: Cooperatively trained conditional and unconditional gan for person image generation. IET Image Processing , 17(10):2949–2957, 2023

    2023

  10. [18]

    Learning transferable visual models from natural language supervision

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. Learning transferable visual models from natural language supervision. In International conference on machine learning , p...

    2021

  11. [19]

    Explor- ing the limits of transfer learning with a unified text-to-text transformer

    Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. Explor- ing the limits of transfer learning with a unified text-to-text transformer. Journal of machine learning research , 21(140):1–67, 2020

    2020

  12. [20]

    Generative multimodal models are in-context learners

    Quan Sun, Yufeng Cui, Xiaosong Zhang, Fan Zhang, Qiying Yu, Yueze Wang, Yongming Rao, Jingjing Liu, Tiejun Huang, and Xinlong Wang. Generative multimodal models are in-context learners. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages...

    2024

  13. [21]

    Attentional generative adversarial networks with representativeness and diversity for generating text to realistic image

    Anjie Tian and Lu Lu. Attentional generative adversarial networks with representativeness and diversity for generating text to realistic image. IEEE access, 8:9587–9596, 2020

    2020

  14. [22]

    Improving the quality of image generation in art with top-k training and cyclic generative methods

    Laura Vela, F ´elix Fuentes-Hurtado, and Adri ´an Colomer. Improving the quality of image generation in art with top-k training and cyclic generative methods. Scientific Reports, 13(1):17764, 2023

    2023

  15. [23]

    Image generation and recognition technology based on attention residual gan

    Huazhe Wang and Li Ma. Image generation and recognition technology based on attention residual gan. IEEE Access, 11:61855–61865, 2023

    2023

  16. [24]

    Image as a foreign language: Beit pretraining for all vision and vision-language tasks

    Wenhui Wang, Hangbo Bao, Li Dong, Johan Bjorck, Zhiliang Peng, Qiang Liu, Kriti Aggarwal, Owais Khan Mohammed, Saksham Singhal, Subhojit Som, et al. Image as a foreign language: Beit pretraining for all vision and vision-language tasks. arXiv preprint arXiv:2208.10442 , 2022

    2022 arXiv

  17. [25]

    Rca-gan: An improved image denoising algorithm based on generative adversarial networks

    Yuming Wang, Shuaili Luo, Liyun Ma, and Min Huang. Rca-gan: An improved image denoising algorithm based on generative adversarial networks. Electronics, 12(22):4595, 2023

    2023

  18. [26]

    Genartist: Multimodal llm as an agent for unified image generation and editing

    Zhenyu Wang, Aoxue Li, Zhenguo Li, and Xihui Liu. Genartist: Multimodal llm as an agent for unified image generation and editing. arXiv preprint arXiv:2407.05600 , 2024

    2024 arXiv

  19. [27]

    Lpgan: A lbp-based proportional input generative adversarial network for image fusion

    Dongxu Yang, Yongbin Zheng, Wanying Xu, Peng Sun, and Di Zhu. Lpgan: A lbp-based proportional input generative adversarial network for image fusion. Remote Sensing, 15(9):2440, 2023

    2023

  20. [28]

    Vsa: Learning varied-size window attention in vision transformers

    Qiming Zhang, Yufei Xu, Jing Zhang, and Dacheng Tao. Vsa: Learning varied-size window attention in vision transformers. In European conference on computer vision , pages 466–483. Springer, 2022

    2022

  21. [29]

    Joint generative image deblurring aided by edge attention prior and dynamic kernel selection

    Zhichao Zhang, Hui Chen, Xiaoqing Yin, and Jinsheng Deng. Joint generative image deblurring aided by edge attention prior and dynamic kernel selection. Wireless Communications and Mobile Computing , 2021(1):1391801, 2021

    2021

  22. [30]

    Cyclic generative attention- adversarial network for low-light image enhancement

    Tong Zhen, Daxin Peng, and Zhihui Li. Cyclic generative attention- adversarial network for low-light image enhancement. Sensors, 23(15):6990, 2023

    2023

  23. [31]

    Adapt or perish: Adaptive sparse transformer with attentive feature refinement for image restoration

    Shihao Zhou, Duosheng Chen, Jinshan Pan, Jinglei Shi, and Jufeng Yang. Adapt or perish: Adaptive sparse transformer with attentive feature refinement for image restoration. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages 2952– 2963, 2024

    2024

  24. [32]

    Wdig: a wavelet domain image generation framework based on frequency domain op- timization

    Qing Zhu, Xiumei Li, Junmei Sun, and Huang Bai. Wdig: a wavelet domain image generation framework based on frequency domain op- timization. EURASIP Journal on Advances in Signal Processing , 2023(1):66, 2023. Sun Weikai He is currently studying for a Bache- lor’s degree in Com...

    2023

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.