Pith. sign in

REVIEW 4 major objections 4 minor 99 references

Reversible Unfolding Network for Concealed Visual Perception with Generative Refinement

T0 review · 4 major / 4 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 RUN++ claims that reversible modeling in both mask and RGB domains, plus a Bernoulli diffusion refiner confined to uncertainty regions, sets new state-of-the-art results in concealed visual perception.

desk verdict A credible but incremental extension of the authors' own RUN that consistently beats its baseline by small margins, with a plausible diffusion-refinement mechanism that the paper never actually tests. read the letter →

arxiv 2508.15027 v1 pith:SFXFU2GI submitted 2025-08-20 cs.CV cs.AIcs.LG

Chunming He , Fengyang Xiao , Rihan Zhang , Chengyu Fang , Deng-Ping Fan , Sina Farsiu This is my paper · ORCID
classification cs.CVcs.AIcs.LG
keywords concealedvisualperceptioncamouflagedobjectdetectiondeepunfoldingnetworkreversiblemodelingdiffusionmodelBernoullisegmentationrefinementbi-leveloptimization
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 is trying to establish that concealed visual perception—detecting objects that blend into their surroundings—improves when reversible foreground/background modeling is applied in both the mask domain and the RGB domain, and when a diffusion model is used only on the pixels where those two reversible views disagree. It formulates the task as an optimization problem and unfolds the iterative solution into a multi-stage deep network, with each stage containing a CORE module for mask-domain reversible modeling, a CARE module for RGB-domain reversible modeling, and a FINE module that runs a targeted Bernoulli diffusion process on uncertain regions. The authors claim this design achieves state-of-the-art results across 14 concealed-object settings, including standard, label-deficient, multimodal, video, and degraded inputs, while keeping computational overhead modest because diffusion is never run on the whole image. A reader should care because the paper offers a principled way to combine model-based optimization, reversible reasoning, and generative refinement for high-level vision tasks that are usually dominated by purely data-driven networks.

What carries the argument

The load-bearing mechanism is the pairing of an uncertainty prior with a restricted diffusion process. CORE and CARE are reversible modules that model the object and its context from two complementary domains, and FINE is a Bernoulli diffusion model—a discrete diffusion process defined directly on binary masks—that acts only on the pixels where CORE and CARE disagree. The unfolding network converts an optimization model for foreground/background separation into a multi-stage deep architecture, with CORE, CARE, and FINE repeated per stage, so that the network supplies a strong uncertainty prior for diffusion rather than generating masks from scratch. The bi-level collaborative optimization fr

What would settle it

A controlled experiment on COD10K where CORE and CARE are made to be systematically wrong in agreement, for example by corrupting their shared encoder so both confidently mark a background patch as foreground. If FINE, restricted to disagreement regions, performs no better than no refinement and clearly worse than a full-image diffusion refiner, the claim that disagreement locates high-uncertainty pixels would fail. Concretely, one could compare disagreement-map-based region targeting against oracle-error-based region targeting on the same segmentation network.

Watch

Extended reading notes

Core claim

The paper's central claim is that concealed-object segmentation is best treated as a foreground/background separation problem solved by alternating two reversible estimators—one in the mask domain (CORE) and one in the RGB domain (CARE)—and then correcting only the pixels where these two estimators disagree. That disagreement region is the paper's uncertainty prior: it is precisely where a targeted Bernoulli diffusion model (FINE) is applied to restore fine detail. The authors formulate the task as a mathematical optimization problem and unfold its iterative solution into a K-stage network, preserving the structure of the optimization while letting all parameters be learned. They also introd

Load-bearing premise

The load-bearing premise is that the pixels where the mask-domain (CORE) and RGB-domain (CARE) modules disagree are exactly the high-uncertainty pixels whose correction improves the mask; if disagreement does not track segmentation error, the diffusion refiner spends its effort on the wrong places.

Editorial extensions

If this is right

  • If the claim holds, full-image diffusion is unnecessary for high-quality CVP refinement; targeting disagreement regions provides most of the benefit at a fraction of the computational cost.
  • Deep unfolding networks, previously used mostly for low-level vision, become a viable and interpretable backbone for high-level tasks like segmentation, because the optimization formulation supplies inductive bias.
  • The bi-level collaborative optimization principle implies that pairing a high-level vision task with a suitable restoration network can improve robustness under real-world degradations, as demonstrated on low-light object detection.
  • Because RUN++ can be integrated with existing segmentation models as a refiner or as a full unfolding backbone, it could serve as a general performance booster rather than only a standalone detector.
  • The method's consistent gains on salient object detection suggest the same two-domain reversible unfolding recipe may transfer to other binary segmentation tasks beyond concealed objects.

Reading between the lines

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

  • A direct test of the disagreement-as-uncertainty heuristic would be to compare CORE/CARE disagreement maps against oracle error maps on COD10K; if disagreement is a weak or biased predictor of true segmentation error, the FINE module's advantage would shrink when compared against a full-image diffusion refiner.
  • The same disagreement prior could be used to schedule compute adaptively per image, spending more diffusion steps on high-disagreement images and fewer on easy ones, an extension the paper does not explore.
  • The paper's stated future direction of probability-oriented models suggests a natural extension: converting the disagreement map into calibrated confidence estimates, which would matter for safety-critical medical or driving applications where knowing when the model is unsure is as important as the mask itself.
  • Because the paper reports that RUN++ improves several existing detectors when integrated into its unfolding structure, a testable prediction is that the same architecture would also boost other dense prediction tasks, such as medical lesion segmentation or defect detection under distribution shift.
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 / 4 minor

Summary. The paper proposes RUN++, a multi-stage deep unfolding network for concealed visual perception (CVP). It claims to formulate CVP as a mathematical optimization problem, unfold the iterative solution into stages containing a CORE module (reversible modeling in the mask domain), a CARE module (reversible modeling in the RGB domain), and a FINE module (a Bernoulli diffusion model that refines only uncertain regions identified by CORE/CARE disagreement). The paper also introduces a degradation-resistant framework and a bi-level collaborative optimization (BLCO) paradigm. Experiments cover standard COD, polyp, tubular, transparent object, defect detection, label-deficient, multimodal, video, salient object detection, and degraded settings, reporting state-of-the-art results across most benchmarks with modest overhead over the authors' prior RUN model.

Significance. The empirical breadth is a clear strength: the paper evaluates across 14 CVP-related settings, multiple backbones, and includes efficiency comparisons and degradation analysis. The targeted-diffusion idea is interesting and computationally motivated. If the central claims hold, the work would provide a general framework unifying reversible modeling, deep unfolding, and generative refinement for CVP. However, the key mechanism—that CORE/CARE disagreement identifies the pixels that diffusion should refine—is not validated, and the unfolding derivation is not fully visible in the reviewed version. These gaps currently prevent the paper from being accepted as a definitive TPAMI contribution.

major comments (4)
  1. [Section III; Table XIX] The abstract and Section III attribute the performance gains to the FINE module's 'strong uncertainty prior' from CORE/CARE disagreement. This mechanism is not validated. Table XIX shows that the FINE-only variants applied to RUN perform at or below the RUN baseline on the reported metrics (e.g., the FINE column is 0.032/0.716/0.896/0.812 versus RUN's 0.031/0.721/0.897/0.812), and the variant using an external uncertainty estimator (FINE1) is no better. Without an analysis of the overlap between the disagreement region and true segmentation errors, or an oracle/random-region baseline, the improvement of RUN++ over RUN could come from added capacity or joint training rather than from targeted diffusion refinement. Please add (i) precision/recall of the disagreement mask against false-positive/false-negative regions, and (ii) ablations refining random regions or ground-truth error regions.
  2. [Section III; Section IV-A] The central 'deep unfolding' claim requires an explicit optimization model and iterative update derivation. In the reviewed version, Section III's equations are not visible—the text jumps from the opening of the methodology to the experimental setup—and Section IV-A states that 'all extra parameters inherited from the original optimization model are treated as learnable variables with random initialization.' If the optimization parameters are randomly initialized and learned, it is unclear what structural inductive bias from the optimization model remains. Please provide the full objective and update equations, and include an ablation that compares the K-stage unfolded architecture against a generic feedforward stack of the same capacity, as well as a study of stage count K, to substantiate that the unfolding structure itself is responsible for the gains.
  3. [Table I; Tables III-VIII; Table XIX] Many of the claimed SOTA margins are very thin, and no error bars, multiple-seed runs, or significance tests are reported. For example, on COD10K in Table I, the S-measure gain of RUN++ over RUN is 0.010 for ResNet50 but only 0.002 for Res2Net50 and 0.005 for PVT V2; several other metrics differ by less than 0.003. The FINE-only comparisons in Table XIX are within 0.001-0.002. Since the paper's primary assertion is state-of-the-art performance, please report standard deviations over at least three runs and, where possible, paired statistical tests on key benchmarks, or explicitly state the evaluation protocol if single-run evaluation is standard in the field.
  4. [Section II] The related-work section describes RUN++ as 'the first deep unfolding network designed for concealed visual perception,' yet the primary comparison baseline, RUN [3], is itself a reversible unfolding network for concealed object segmentation. This claim is internally inconsistent and should be rephrased. More importantly, the incremental contribution over RUN should be made explicit: the paper should isolate the effect of the CARE module, the FINE module, and the degradation-resistant extension through component-wise ablations, rather than only comparing the full RUN++ against RUN.
minor comments (4)
  1. [Table XXI] The table header contains a typo: 'Restuls' should be 'Results'.
  2. [Section IV-A] The sentence 'only set FINE in the last stage' should be reworded as 'set FINE only in the last stage.' Also, the cosine noise schedule text has a placeholder where the number of timesteps should appear.
  3. [Section IV-B] Several numerical gains are rendered as placeholder glyphs (e.g., 'average performance gains of �����'). The final version must ensure all numbers are visible and correct.
  4. [Table XIX] The suffix definition says methods with '+' are retrained, but the FINE and FINE1 columns do not have a '+' suffix. Please clarify whether FINE and FINE1 were retrained as part of the comparison or applied in a frozen/off-the-shelf manner.

Circularity Check

0 steps flagged · score 1.0 of 10

No circular derivation found; RUN++ is an incremental extension of the authors' RUN with external benchmarks, and the FINE uncertainty prior, while empirically unvalidated, is not a by-construction reduction.

full rationale

The paper's derivation chain is: (1) define a foreground-background reversible optimization objective, (2) unfold its iterative solution into multi-stage modules CORE/CARE, (3) use CORE/CARE prediction disagreement as a spatial prior for a Bernoulli diffusion refiner FINE, and (4) validate against external baselines across 14 CVP settings. None of these steps reduces to its own inputs by construction. The CORE/CARE disagreement is an internal network signal, not a fitted constant, and FINE is trained with ground-truth mask supervision rather than being defined to equal the disagreement map. The claimed SOTA results are measured against non-self baselines such as FocusDiff, CamoDiff, SegRefiner, and BiRefNet on held-out test sets, so the central empirical claim is self-contained. The paper does reuse the authors' own RUN, FEDER, and UnfoldIR as components and baselines, but this is incremental reuse rather than load-bearing circularity: RUN++ is tested against RUN itself, and Table XIX ablates the uncertainty prior against external estimators (UGTR and entropy), so the contribution is not assumed via self-citation. The skeptic's concern that the CORE/CARE disagreement region may not track true segmentation errors is a legitimate empirical weakness and a missing validation, but it is not a circular step because the paper does not define the improvement to be equivalent to the disagreement. Therefore, under the hard rules requiring a specific reduction by equation or by definition, no circularity step is established.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim rests on the modeling assumption that CVP is a foreground-background separation problem with reversible priors, on the assumption that CORE and CARE disagreement marks uncertainty, and on the standard DUN assumption that a finite unfolding approximates the optimization, whose parameters are fitted rather than derived. No new physical entities are introduced. The method's inputs are standard datasets; its outputs are segmentation masks.

free parameters (4)
  • Optimization-model parameters (step sizes, regularization weights, uncertainty threshold) = learned end-to-end from random initialization
    Section IV-A: 'All extra parameters inherited from the original optimization model are treated as learnable variables with random initialization.'
  • Number of unfolding stages K = 4 (RUN++-4)
    Table I reports RUN++-4; chosen by hand as the multi-stage setting.
  • DDIM sampling steps = 5
    Section IV-A: 'we utilized DDIM sampling with 5 timesteps'. Hand-chosen inference setting.
  • Diffusion training timesteps and noise schedule = cosine schedule, 1000 timesteps
    Section IV-A: 'a cosine noise schedule with 1000 timesteps during training'. Hand-chosen hyperparameters.
assumptions (5)
  • domain assumption CVP can be modeled as foreground-background separation under a reversible (complementary) prior.
    The optimization model in Section III is constructed by the authors; no derivation shows this prior is necessary or sufficient for concealed object segmentation.
  • ad hoc to paper Disagreement between the CORE and CARE module outputs identifies high-uncertainty regions worthy of diffusion refinement.
    The FINE module's uncertainty prior is defined as the difference between CORE and CARE outputs; the paper provides no independent evidence that disagreement correlates with segmentation error.
  • ad hoc to paper A K-stage unfolding network approximates the fixed point of the iterative optimization, with all optimization-derived parameters learnable.
    Implementation section IV-A states optimization parameters are learnable with random initialization; no convergence guarantee or bound links the network stages to the optimization solution.
  • standard math Multinomial and Bernoulli diffusion forward processes for discrete masks are valid generative models.
    Borrowed from Hoogeboom et al. (Ref 24) and Sohl-Dickstein et al. (Ref 20); used for the FINE refiner.
  • domain assumption Replacing the internal reconstruction network with SOTA restoration networks (CORUN, UnfoldIR, DiffIR) improves downstream segmentation under degradation.
    Section III-D and Fig. 10 validate this empirically, but it is not derived; it is an ensemble hypothesis whose success depends on complementarity of the restoration task.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Reversible Unfolding Network for Concealed Visual Perception with Generative Refinement." pith.science (2026). https://pith.science/paper/SFXFU2GI

@misc{pith2026250815027,
  author       = {Pith},
  title        = {Pith review of: Reversible Unfolding Network for Concealed Visual Perception with Generative Refinement},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SFXFU2GI}},
  note         = {Machine review of arXiv:2508.15027}
}
read the original abstract

Existing methods for concealed visual perception (CVP) often leverage reversible strategies to decrease uncertainty, yet these are typically confined to the mask domain, leaving the potential of the RGB domain underexplored. To address this, we propose a reversible unfolding network with generative refinement, termed RUN++. Specifically, RUN++ first formulates the CVP task as a mathematical optimization problem and unfolds the iterative solution into a multi-stage deep network. This approach provides a principled way to apply reversible modeling across both mask and RGB domains while leveraging a diffusion model to resolve the resulting uncertainty. Each stage of the network integrates three purpose-driven modules: a Concealed Object Region Extraction (CORE) module applies reversible modeling to the mask domain to identify core object regions; a Context-Aware Region Enhancement (CARE) module extends this principle to the RGB domain to foster better foreground-background separation; and a Finetuning Iteration via Noise-based Enhancement (FINE) module provides a final refinement. The FINE module introduces a targeted Bernoulli diffusion model that refines only the uncertain regions of the segmentation mask, harnessing the generative power of diffusion for fine-detail restoration without the prohibitive computational cost of a full-image process. This unique synergy, where the unfolding network provides a strong uncertainty prior for the diffusion model, allows RUN++ to efficiently direct its focus toward ambiguous areas, significantly mitigating false positives and negatives. Furthermore, we introduce a new paradigm for building robust CVP systems that remain effective under real-world degradations and extend this concept into a broader bi-level optimization framework.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

99 extracted references · 76 canonical work pages

  1. [3]

    Run: Reversible unfolding network for concealed object segmentation,

    C. He, R. Zhang, F. Xiao, C. Fang, L. Tang, Y . Zhang, L. Kong, D.- P. Fan, K. Li, and S. Farsiu, “Run: Reversible unfolding network for concealed object segmentation,” ICML, 2025. 1, 2, 3, 9, 10, 11, 12, 14, 15

    2025

  2. [1]

    Focus- diffuser: Perceiving local disparities for camouflaged object detection,

    J. Zhao, X. Li, F. Yang, Q. Zhai, A. Luo, Z. Jiao, and H. Cheng, “Focus- diffuser: Perceiving local disparities for camouflaged object detection,” in ECCV, 2024, pp. 181–198. 2, 3, 9

    2024

  3. [2]

    Conditional diffusion models for camouflaged and salient object detection,

    K. Sun, Z. Chen, X. Lin, X. Sun, H. Liu, and R. Ji, “Conditional diffusion models for camouflaged and salient object detection,” IEEE Transactions on Pattern Analysis and Machine Intelligence , 2025. 2, 3, 9

    2025

  4. [4]

    Camouflaged object detection,

    D.-P. Fan, G.-P. Ji, G. Sun, M.-M. Cheng, and J. Shen, “Camouflaged object detection,” in CVPR, 2020, pp. 2777–2787. 1, 8

    2020

  5. [5]

    Weakly-supervised concealed object segmentation with sam-based pseudo labeling and multi-scale feature grouping,

    C. He, K. Li, Y . Zhang, G. Xu, and L. Tang, “Weakly-supervised concealed object segmentation with sam-based pseudo labeling and multi-scale feature grouping,” NeurIPS, 2023. 1, 3, 4, 6, 11

    2023

  6. [6]

    Strategic preys make acute predators: Enhancing camouflaged object detectors by generating camouflaged objects,

    C. He, K. Li, Y . Zhang, Y . Zhang, Z. Guo, and X. Li, “Strategic preys make acute predators: Enhancing camouflaged object detectors by generating camouflaged objects,” ICLR, 2024. 1

    2024

  7. [7]

    Image threshold segmentation based on glle histogram,

    X. Wang, L. Deng, and G. Xu, “Image threshold segmentation based on glle histogram,” in CPSCom. IEEE, 2019, pp. 410–415. 1

    2019

  8. [8]

    Camouflaged object detection with feature decomposition and edge reconstruction,

    C. He, K. Li, Y . Zhang, L. Tang, and Y . Zhang, “Camouflaged object detection with feature decomposition and edge reconstruction,” inCVPR, 2023, pp. 22 046–22 055. 1, 3, 6, 7, 8, 9, 10, 13, 15

    2023

Show all 99 references
  1. [9]

    A survey of camouflaged object detection and beyond,

    F. Xiao, S. Hu, Y . Shen, and C. He, “A survey of camouflaged object detection and beyond,” CAAI AIR, 2024. 3

    2024

  2. [10]

    Concealed object segmentation with hierarchical coherence modeling,

    F. Xiao, P. Zhang, and C. He, “Concealed object segmentation with hierarchical coherence modeling,” in CAAI ICAI, 2023, pp. 16–27. 3

    2023

  3. [11]

    Pranet: Parallel reverse attention network for polyp segmentation,

    D.-P. Fan, G.-P. Ji, and T. Zhou, “Pranet: Parallel reverse attention network for polyp segmentation,” in MICCAI, 2020, pp. 263–273. 3, 9

    2020

  4. [12]

    Bilateral reference for high-resolution dichotomous image segmentation,

    P. Zheng, D. Gao, D.-P. Fan, L. Liu, J. Laaksonen, W. Ouyang, and N. Sebe, “Bilateral reference for high-resolution dichotomous image segmentation,” CAAI AIR, 2024. 3, 9, 10

    2024

  5. [13]

    Segrefiner: Towards model-agnostic segmentation refinement with discrete diffusion process,

    M. Wang, H. Ding, J. H. Liew, J. Liu, Y . Zhao, and Y . Wei, “Segrefiner: Towards model-agnostic segmentation refinement with discrete diffusion process,” NeurIPS, 2023. 3, 13, 14

    2023

  6. [14]

    Degradation-resistant unfolding network for heterogeneous image fusion,

    C. He, K. Li, and Y . Zhang, “Degradation-resistant unfolding network for heterogeneous image fusion,” in ICCV, 2023, pp. 611–621. 3, 13, 14

    2023

  7. [15]

    Real-world image dehazing with coherence-based pseudo labeling and cooperative unfolding network,

    C. Fang, C. He, F. Xiao, Y . Zhang, L. Tang, Y . Zhang, K. Li, and X. Li, “Real-world image dehazing with coherence-based pseudo labeling and cooperative unfolding network,” in NeurIPS, 2024. 3, 5, 14

    2024

  8. [16]

    Optimization of lipschitz continuous functions,

    A. A. Goldstein, “Optimization of lipschitz continuous functions,” Math. Program., vol. 13, pp. 14–22, 1977. 5

    1977

  9. [17]

    Vmamba: Visual state space model,

    Y . Liu, Y . Tian, Y . Zhao, H. Yu, L. Xie, Y . Wang, Q. Ye, and Y . Liu, “Vmamba: Visual state space model,” in NeurIPS, 2024. 5

    2024

  10. [18]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in CVPR, 2016, pp. 770–778. 6, 10

    2016

  11. [19]

    Hqg-net: Unpaired medical image enhancement with high-quality guidance,

    G. Xu, J. Yan, L. Tang, and Y . Zhang, “Hqg-net: Unpaired medical image enhancement with high-quality guidance,” IEEE Trans. Neural Networks Learn. Syst. , 2023. 6, 8, 13, 14

    2023

  12. [20]

    Deep unsupervised learning using nonequilibrium thermodynamics,

    J. Sohl-Dickstein, E. Weiss, N. Maheswaranathan, and S. Ganguli, “Deep unsupervised learning using nonequilibrium thermodynamics,” in ICML, 2015, pp. 2256–2265. 6

    2015

  13. [21]

    Segment concealed object with incomplete supervision,

    C. He, K. Li, Y . Zhang, Z. Yang, L. Tang, Y . Zhang, L. Kong, and S. Farsiu, “Segment concealed object with incomplete supervision,” IEEE Trans. Pattern Anal. Mach. Intell. , 2025. 6, 11, 13

    2025

  14. [22]

    Uncertainty-guided transformer reasoning for camouflaged object detection,

    F. Yang, Q. Zhai, X. Li, R. Huang, A. Luo, H. Cheng, and D.-P. Fan, “Uncertainty-guided transformer reasoning for camouflaged object detection,” in ICCV, 2021, pp. 4146–4155. 6, 13, 14

    2021

  15. [23]

    Auto-encoding variational bayes,

    D. P. Kingma, M. Welling et al. , “Auto-encoding variational bayes,”

  16. [24]

    Argmax flows and multinomial diffusion: Learning categorical distributions,

    E. Hoogeboom, D. Nielsen, P. Jaini, P. Forr’e, and M. Welling, “Argmax flows and multinomial diffusion: Learning categorical distributions,” in NeurIPS, 2021. 7

    2021

  17. [25]

    Reti-diff: Illumination degradation image restoration with retinex-based latent diffusion model,

    C. He, C. Fang, Y . Zhang, T. Ye, K. Li, L. Tang, Z. Guo, X. Li, and S. Farsiu, “Reti-diff: Illumination degradation image restoration with retinex-based latent diffusion model,” ICLR, 2025. 7, 13, 14, 15

    2025

  18. [26]

    Iqpfr: An image quality prior for blind face restoration and beyond,

    P. Hu, C. He, L. Xu, J. Tian, S. Farsiu, Y . Zhang, P. Liu, and X. Li, “Iqpfr: An image quality prior for blind face restoration and beyond,” arXiv preprint arXiv:2503.09294 , 2025. 7

    2025 arXiv

  19. [27]

    Diffir: Efficient diffusion model for image restoration,

    B. Xia, Y . Zhang, S. Wang, Y . Wang, X. Wu, Y . Tian, W. Yang, and L. Van Gool, “Diffir: Efficient diffusion model for image restoration,” in ICCV, 2023, pp. 13 095–13 105. 7, 14

    2023

  20. [28]

    Simultaneously localize, segment and rank the camouflaged objects,

    Y . Lv, J. Zhang, Y . Dai, A. Li, B. Liu, N. Barnes, and D.-P. Fan, “Simultaneously localize, segment and rank the camouflaged objects,” in CVPR, 2021, pp. 11 591–11 601. 9

    2021

  21. [29]

    Exploring figure-ground assignment mechanism in perceptual organization,

    W. Zhai, Y . Cao, and J. Zhang, “Exploring figure-ground assignment mechanism in perceptual organization,” in NeurIPS, vol. 35, 2023. 9

    2023

  22. [30]

    Frequency-spatial entanglement learning for camouflaged object detection,

    Y . Sun, C. Xu, J. Yang, H. Xuan, and L. Luo, “Frequency-spatial entanglement learning for camouflaged object detection,” in ECCV, 2024, pp. 343–360. 9

    2024

  23. [31]

    I can find you! boundary- guided separated attention network for camouflaged object detection,

    H. Zhu, P. Li, H. Xie, X. Yan, and D. Liang, “I can find you! boundary- guided separated attention network for camouflaged object detection,” in AAAI, vol. 36, 2022, pp. 3608–3616. 9

    2022

  24. [32]

    Camofocus: En- hancing camouflage object detection with split-feature focal modulation and context refinement,

    A. Khan, M. Khan, W. Gueaieb, and A. El Saddik, “Camofocus: En- hancing camouflage object detection with split-feature focal modulation and context refinement,” in WACV, 2024, pp. 1434–1443. 9

    2024

  25. [33]

    Zoom in and out: A mixed-scale triplet network for camouflaged object detection,

    Y . Pang, X. Zhao, T.-Z. Xiang, L. Zhang, and H. Lu, “Zoom in and out: A mixed-scale triplet network for camouflaged object detection,” in CVPR, 2022, pp. 2160–2170. 9, 10

    2022

  26. [34]

    Zoomnext: A unified collaborative pyramid network for camou- flaged object detection,

    ——, “Zoomnext: A unified collaborative pyramid network for camou- flaged object detection,” IEEE Trans. Pattern Anal. Mach. Intell. , 2024. 9, 10, 12

    2024

  27. [35]

    Segment, magnify and reiterate: Detect camouflaged objects hard way,

    Q. Jia, S. Yao, and Y . Liu, “Segment, magnify and reiterate: Detect camouflaged objects hard way,” in CVPR, 2022, pp. 713–722. 9, 10

    2022

  28. [36]

    Target- aware dual adversarial learning and a multi-scenario multi-modality benchmark to fuse infrared and visible for object detection,

    J. Liu, X. Fan, Z. Huang, G. Wu, R. Liu, W. Zhong, and Z. Luo, “Target- aware dual adversarial learning and a multi-scenario multi-modality benchmark to fuse infrared and visible for object detection,” in CVPR, 2022, pp. 5802–5811. 8

    2022

  29. [37]

    Bilevel optimization with nonsmooth lower level problems,

    P. Ochs, R. Ranftl, T. Brox, and T. Pock, “Bilevel optimization with nonsmooth lower level problems,” in SSVM. Springer, 2015, pp. 654–

    2015

  30. [38]

    Medical image segmentation via cascaded attention decoding,

    M. M. Rahman, “Medical image segmentation via cascaded attention decoding,” in WACV, 2023, pp. 6222–6231. 9

    2023

  31. [39]

    Polyp-pvt: Polyp segmentation with pyramid vision transformers,

    B. Dong, W. Wang, D.-P. Fan, J. Li, and H. Fu, “Polyp-pvt: Polyp segmentation with pyramid vision transformers,” CAAI AIR , 2023. 9, 10

    2023

  32. [40]

    Coinnet: A convolution- involution network with a novel statistical attention for automatic polyp segmentation,

    S. Jain, R. Atale, A. Gupta, and U. Mishra, “Coinnet: A convolution- involution network with a novel statistical attention for automatic polyp segmentation,” IEEE Trans. Med. Imaging , pp. 3987–4000, 2023. 9

    2023

  33. [41]

    Lssnet: A method for colon polyp segmentation based on local feature supplementation and shallow feature supplementation,

    W. Wang, H. Sun, and X. Wang, “Lssnet: A method for colon polyp segmentation based on local feature supplementation and shallow feature supplementation,” in MICCAI. Springer, 2024, pp. 446–456. 9, 10

    2024

  34. [42]

    Cs2-net: Deep learning segmentation of curvilinear structures in medical imaging,

    L. Mou, Y . Zhao, H. Fu, Y . Liu, J. Cheng, Y . Zheng, P. Su, and J. Yang, “Cs2-net: Deep learning segmentation of curvilinear structures in medical imaging,” Med. Image Anal. , vol. 67, p. 101874, 2021. 9

    2021

  35. [43]

    Dynamic snake convo- lution based on topological geometric constraints for tubular structure segmentation,

    Y . Qi, Y . He, X. Qi, Y . Zhang, and G. Yang, “Dynamic snake convo- lution based on topological geometric constraints for tubular structure segmentation,” in ICCV, 2023, pp. 6070–6079. 9, 10

    2023

  36. [44]

    Stimulus-guided adaptive transformer network for retinal blood vessel segmentation in fundus images,

    J. Lin, X. Huang, H. Zhou, Y . Wang, and Q. Zhang, “Stimulus-guided adaptive transformer network for retinal blood vessel segmentation in fundus images,” Med. Image Anal. , vol. 89, p. 102929, 2023. 9 SUBMITTED TO IEEE TPAMI 17

    2023

  37. [45]

    Topology-aware uncertainty for image segmentation,

    S. Gupta, Y . Zhang, X. Hu, P. Prasanna, and C. Chen, “Topology-aware uncertainty for image segmentation,” NeurIPS, vol. 36, 2024. 9

    2024

  38. [46]

    Represent- ing topological self-similarity using fractal feature maps for accurate segmentation of tubular structures,

    J. Huang, Y . Zhou, Y . Luo, G. Liu, H. Guo, and G. Yang, “Represent- ing topological self-similarity using fractal feature maps for accurate segmentation of tubular structures,” in ECCV, 2025, pp. 143–160. 9

    2025

  39. [47]

    Animal camouflage analysis: Chameleon database,

    P. Skurowski, H. Abdulameer, and J. Błaszczyk, “Animal camouflage analysis: Chameleon database,” Unpublished manuscript, p. 7, 2018. 9

    2018

  40. [48]

    Anabranch network for camouflaged object segmentation,

    T.-N. Le, T. V . Nguyen, Z. Nie, M.-T. Tran, and A. Sugimoto, “Anabranch network for camouflaged object segmentation,”Comput. Vis. Image Underst., vol. 184, pp. 45–56, 2019. 9

    2019

  41. [49]

    Concealed object detection,

    D.-P. Fan, G.-P. Ji, M.-M. Cheng, and L. Shao, “Concealed object detection,” IEEE Trans. Pattern Anal. Mach. Intell. , 2021. 9, 10, 12

    2021

  42. [50]

    How to evaluate foreground maps?

    R. Margolin, L. Zelnik-Manor, and A. Tal, “How to evaluate foreground maps?” in CVPR, 2014, pp. 248–255. 9

    2014

  43. [51]

    Cognitive vision inspired object segmen- tation metric and loss function,

    D.-P. Fan, G. Ji, and X. Qin, “Cognitive vision inspired object segmen- tation metric and loss function,” SCI. SIN. Informationis , 2021. 10

    2021

  44. [52]

    Structure-measure: A new way to evaluate foreground maps,

    D.-P. Fan, M.-M. Cheng, Y . Liu, and T. Li, “Structure-measure: A new way to evaluate foreground maps,” in ICCV, 2017, pp. 4548–4557. 10

    2017

  45. [53]

    Res2net: A new multi-scale backbone architecture,

    S.-H. Gao, M.-M. Cheng, K. Zhao, X.-Y . Zhang, M.-H. Yang, and P. Torr, “Res2net: A new multi-scale backbone architecture,” IEEE Trans. Pattern Anal. Mach. Intell. , vol. 43, no. 2, pp. 652–662, 2019. 10

    2019

  46. [54]

    Pvt v2: Improved baselines with pyramid vision transformer,

    W. Wang, E. Xie, X. Li, D.-P. Fan, K. Song, D. Liang, T. Lu, P. Luo, and L. Shao, “Pvt v2: Improved baselines with pyramid vision transformer,” Comput. Vis. Media , vol. 8, no. 3, pp. 415–424, 2022. 10

    2022

  47. [55]

    Don’t hit me! glass detection in real-world scenes,

    H. Mei, X. Yang, Y . Wang, Y . Liu, and S. He, “Don’t hit me! glass detection in real-world scenes,” in CVPR, 2020, pp. 3687–3696. 10

    2020

  48. [56]

    Enhanced boundary learning for glass-like object segmen- tation,

    H. He, X. Li, G. Cheng, J. Shi, Y . Tong, G. Meng, V . Prinet, and L. Weng, “Enhanced boundary learning for glass-like object segmen- tation,” in ICCV, 2021, pp. 15 859–15 868. 10

    2021

  49. [57]

    Rfenet: towards reciprocal feature evolution for glass segmentation,

    K. Fan, C. Wang, and Y . Wang, “Rfenet: towards reciprocal feature evolution for glass segmentation,” in IJCAI, 2023, pp. 717–725. 10

    2023

  50. [58]

    Internal-external boundary attention fusion for glass surface segmentation,

    D. Han, S. Lee, C. Zhang, H. Yoon, H. Kwon, H.-C. Kim, and H.- G. Choo, “Internal-external boundary attention fusion for glass surface segmentation,” arXiv preprint arXiv:2307.00212 , 2024. 10

    2024 arXiv

  51. [59]

    Ghostingnet: A novel approach for glass surface detection with ghosting cues,

    T. Yan, J. Gao, K. Xu, X. Zhu, H. Huang, H. Li, B. Wah, and R. W. Lau, “Ghostingnet: A novel approach for glass surface detection with ghosting cues,” IEEE Trans. Pattern Anal. Mach. Intell. , 2024. 10

    2024

  52. [60]

    Automated polyp detection in colonoscopy videos using shape and context information,

    N. Tajbakhsh, S. R. Gurudu, and J. Liang, “Automated polyp detection in colonoscopy videos using shape and context information,” IEEE Trans. Med. Imaging, vol. 35, no. 2, pp. 630–644, 2015. 10

    2015

  53. [61]

    Toward embedded detection of polyps in wce images for early diagnosis,

    J. Silva, A. Histace, O. Romain, and X. Dray, “Toward embedded detection of polyps in wce images for early diagnosis,” Int. J. Comput. Assist. Radiol. Surg. , vol. 9, pp. 283–293, 2014. 10

    2014

  54. [62]

    Struc- ture and illumination constrained gan for medical image enhancement,

    Y . Ma, J. Liu, Y . Liu, H. Fu, Y . Hu, J. Cheng, H. Qi, and Y . Zhao, “Struc- ture and illumination constrained gan for medical image enhancement,” IEEE Trans. Med. Imaging , pp. 3955–3967, 2021. 10

    2021

  55. [63]

    High-resolution iterative feedback network for camouflaged object detection,

    X. Hu, S. Wang, X. Qin, H. Dai, W. Ren, D. Luo, Y . Tai, and L. Shao, “High-resolution iterative feedback network for camouflaged object detection,” in AAAI, vol. 37, no. 1, 2023, pp. 881–889. 10

    2023

  56. [64]

    Camoformer: Masked separable attention for camouflaged object detection,

    B. Yin, X. Zhang, D.-P. Fan, S. Jiao, M.-M. Cheng, L. Van Gool, and Q. Hou, “Camoformer: Masked separable attention for camouflaged object detection,” IEEE Trans. Pattern Anal. Mach. Intell. , 2024. 10

    2024

  57. [65]

    Oaformer: Occlusion aware transformer for camouflaged object detection,

    X. Yang, H. Zhu, and G. Mao, “Oaformer: Occlusion aware transformer for camouflaged object detection,” in ICME, 2023, pp. 1421–1426. 10

    2023

  58. [66]

    Sam-adapter: Adapting segment anything in underperformed scenes,

    T. Chen, L. Zhu, C. Deng, R. Cao, Y . Wang, and S. Zhang, “Sam-adapter: Adapting segment anything in underperformed scenes,” in ICCV, 2023, pp. 3367–3375. 11

    2023

  59. [67]

    Weakly-supervised camouflaged object detection with scribble,

    R. He, Q. Dong, and J. Lin, “Weakly-supervised camouflaged object detection with scribble,” AAAI, 2023. 11

    2023

  60. [68]

    Relax image-specific prompt requirement in sam: A single generic prompt for segmenting camouflaged objects,

    J. Hu, J. Lin, and S. Gong, “Relax image-specific prompt requirement in sam: A single generic prompt for segmenting camouflaged objects,” in AAAI, vol. 38, no. 11, 2024, pp. 12 511–12 518. 11

    2024

  61. [69]

    Pseudo-label guided contrastive learning for semi- supervised medical image segmentation,

    H. Basak and Z. Yin, “Pseudo-label guided contrastive learning for semi- supervised medical image segmentation,” in CVPR, 2023, pp. 19 786– 19 797. 11

    2023

  62. [70]

    Saliency as pseudo-pixel supervision for weakly and semi-supervised semantic segmentation,

    M. Lee, S. Lee, J. Lee, and H. Shim, “Saliency as pseudo-pixel supervision for weakly and semi-supervised semantic segmentation,” IEEE Trans. Pattern Anal. Mach. Intell. , vol. 45, no. 10, pp. 12 341– 12 357, 2023. 11

    2023

  63. [71]

    Unsupervised and semi-supervised co-salient object detection via segmentation fre- quency statistics,

    S. Chakraborty, S. Naha, M. Bastan, D. Samaras et al., “Unsupervised and semi-supervised co-salient object detection via segmentation fre- quency statistics,” in WACV, 2024, pp. 332–342. 11

    2024

  64. [72]

    Source-free depth for object pop-out,

    Z. Wu, D. P. Paudel, D.-P. Fan, J. Wang, S. Wang, C. Demonceaux, R. Timofte, and L. Van Gool, “Source-free depth for object pop-out,” in ICCV, 2023, pp. 1032–1042. 11, 12

    2023

  65. [73]

    Integrating extra modality helps segmentor find camouflaged objects well,

    C. Fang, C. He, L. Tang, Y . Zhang, C. Zhu, Y . Shen, C. Chen, G. Xu, and X. Li, “Integrating extra modality helps segmentor find camouflaged objects well,” arXiv preprint arXiv:2502.14471 , 2025. 11, 12

    2025 arXiv

  66. [74]

    Specificity- preserving rgb-d saliency detection,

    T. Zhou, H. Fu, G. Chen, Y . Zhou, D.-P. Fan, and L. Shao, “Specificity- preserving rgb-d saliency detection,” in ICCV, 2021, pp. 4681–4691. 11

    2021

  67. [75]

    Spsn: Superpixel prototype sam- pling network for rgb-d salient object detection,

    M. Lee, C. Park, S. Cho, and S. Lee, “Spsn: Superpixel prototype sam- pling network for rgb-d salient object detection,” in ECCV. Springer, 2022, pp. 630–647. 11

    2022

  68. [76]

    Exploring deeper! segment anything model with depth perception for camouflaged object detection,

    Z. Yu, X. Zhang, L. Zhao, Y . Bin, and G. Xiao, “Exploring deeper! segment anything model with depth perception for camouflaged object detection,” in ACM MM, 2024, pp. 4322–4330. 11

    2024

  69. [77]

    Advances in deep concealed scene understanding,

    D.-P. Fan, G.-P. Ji, P. Xu, M.-M. Cheng, and L. Van Gool, “Advances in deep concealed scene understanding,” Visual Intell., p. 16, 2023. 11

    2023

  70. [78]

    Progressively normalized self-attention network for video polyp seg- mentation,

    G.-P. Ji, Y .-C. Chou, D.-P. Fan, G. Chen, H. Fu, D. Jha, and L. Shao, “Progressively normalized self-attention network for video polyp seg- mentation,” in MICCAI, 2021, pp. 142–152. 12

    2021

  71. [79]

    Self- supervised video object segmentation by motion grouping,

    C. Yang, H. Lamdouar, E. Lu, A. Zisserman, and W. Xie, “Self- supervised video object segmentation by motion grouping,” in ICCV, 2021, pp. 7177–7188. 12

    2021

  72. [80]

    Implicit motion handling for video camouflaged object detection,

    X. Cheng, H. Xiong, D.-P. Fan, Y . Zhong, M. Harandi, T. Drummond, and Z. Ge, “Implicit motion handling for video camouflaged object detection,” in CVPR, 2022, pp. 13 864–13 873. 12

    2022

  73. [81]

    It’s moving! a probabilistic model for causal motion segmentation in moving camera videos,

    P. Bideau and E. Learned-Miller, “It’s moving! a probabilistic model for causal motion segmentation in moving camera videos,” in ECCV, 2016, pp. 433–449. 12

    2016

  74. [82]

    Efficient inference in fully connected crfs with gaussian edge potentials,

    P. Kr ¨ahenb¨uhl and V . Koltun, “Efficient inference in fully connected crfs with gaussian edge potentials,” NeurIPS, vol. 24, 2011. 13, 14

    2011

  75. [83]

    The fast bilateral solver,

    J. T. Barron and B. Poole, “The fast bilateral solver,” in ECCV. Springer, 2016, pp. 617–632. 13, 14

    2016

  76. [84]

    Samrefiner: Taming segment anything model for universal mask refinement,

    Y . Lin, H. Li, W. Shao, Z. Yang, J. Zhao, X. He, P. Luo, and K. Zhang, “Samrefiner: Taming segment anything model for universal mask refinement,” ICLR, 2025. 13, 14

    2025

  77. [85]

    Visual saliency transformer,

    N. Liu, N. Zhang, K. Wan, L. Shao, and J. Han, “Visual saliency transformer,” in ICCV, 2021, pp. 4722–4732. 15

    2021

  78. [86]

    Salient object detection via integrity learning,

    M. Zhuge, D.-P. Fan, N. Liu, and D. Zhang, “Salient object detection via integrity learning,” IEEE Trans. Pattern Anal. Mach. Intell. , 2022. 15

    2022

  79. [87]

    Pyramid grafting network for one-stage high resolution saliency detection,

    C. Xie, C. Xia, and M. Ma, “Pyramid grafting network for one-stage high resolution saliency detection,” in CVPR, 2022, pp. 717–726. 15

    2022

  80. [88]

    Pixels, regions, and objects: Multiple enhancement for salient object detection,

    Y . Wang, R. Wang, and X. Fan, “Pixels, regions, and objects: Multiple enhancement for salient object detection,” in CVPR, 2023, pp. 31–40. 15

    2023

  81. [89]

    Recurrent multi-scale transformer for salient object detection,

    X. Deng, P. Zhang, and W. Liu, “Recurrent multi-scale transformer for salient object detection,” in ACM MM, 2023, pp. 413–423. 15

    2023

  82. [90]

    Gponet: A two-stream gated progressive optimization network for salient object detection,

    Y . Yi, N. Zhang, W. Zhou, Y . Shi, G. Xie, and J. Wang, “Gponet: A two-stream gated progressive optimization network for salient object detection,” Pattern Recogn., vol. 150, p. 110330, 2024. 15

    2024

  83. [91]

    Efficient and stronger visual saliency transformer,

    N. Liu, Z. Luo, N. Zhang, and J. Han, “Efficient and stronger visual saliency transformer,” IEEE Trans. Pattern Anal. Mach. Intell. , 2024. 15

    2024

  84. [92]

    Unfoldir: Rethinking deep unfolding network in illumination degrada- tion image restoration,

    C. He, R. Zhang, F. Xiao, C. Fang, L. Tang, Y . Zhang, and S. Farsiu, “Unfoldir: Rethinking deep unfolding network in illumination degrada- tion image restoration,” arXiv preprint arXiv:2505.06683 , 2025. 14, 15

    2025 arXiv

  85. [93]

    Getting to know low-light images with the exclusively dark dataset,

    Y . P. Loh and C. S. Chan, “Getting to know low-light images with the exclusively dark dataset,” Computer Vision and Image Understanding , vol. 178, pp. 30–42, 2019. 15

    2019

  86. [94]

    Diff-retinex: Rethinking low-light image enhancement with a generative diffusion model,

    X. Yi, H. Xu, H. Zhang, L. Tang, and J. Ma, “Diff-retinex: Rethinking low-light image enhancement with a generative diffusion model,” in ICCV, 2023, pp. 12 302–12 311. 15

    2023

  87. [95]

    Saliency detection via graph-based manifold ranking,

    C. Yang, L. Zhang, H. Lu, X. Ruan, and M.-H. Yang, “Saliency detection via graph-based manifold ranking,” in CVPR, 2013, pp. 3166–3173. 15

    2013

  88. [96]

    Learning to detect salient objects with image-level supervision,

    L. Wang, H. Lu, and Y . Wang, “Learning to detect salient objects with image-level supervision,” in CVPR, 2017, pp. 136–145. 15

    2017

  89. [97]

    Hierarchical saliency detection,

    Q. Yan, L. Xu, J. Shi, and J. Jia, “Hierarchical saliency detection,” in CVPR, 2013, pp. 1155–1162. 15

    2013

  90. [98]

    Visual saliency based on multiscale deep features,

    G. Li and Y . Yu, “Visual saliency based on multiscale deep features,” in CVPR, 2015, pp. 5455–5463. 15

    2015

  91. [99]

    The secrets of salient object segmentation,

    Y . Li, X. Hou, C. Koch, J. M. Rehg, and A. L. Yuille, “The secrets of salient object segmentation,” in CVPR, 2014, pp. 280–287. 15 Chunming He received the B.S. degree from Nanjing University of Posts and Telecommunications, China, and the M.E. degree from Tsinghua University...

    2014

Pith tools

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