pith. sign in

arxiv: 2605.15450 · v1 · pith:QMGYHQKEnew · submitted 2026-05-14 · 💻 cs.CV · cs.AI· cs.LG

RIDE: Retinex-Informed Decoupling for Exposing Concealed Objects

Chunming He , Rihan Zhang , Dingming Zhang , Chengyu Fang , Longxiang Tang , Jingjia Feng , Fengyang Xiao , Sina Farsiu This is my paper

Pith reviewed 2026-05-19 15:04 UTC · model grok-4.3

classification 💻 cs.CV cs.AIcs.LG
keywords Retinex decompositionConcealed object segmentationCamouflaged object detectionDiscriminability Gap TheoremImage decompositionPolyp segmentationTransparent object detectionIndustrial defect inspection
0
0 comments X

The pith

Retinex decomposition separates illumination from reflectance to widen the gap between foreground and background in concealed object segmentation.

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

The paper argues that in tasks such as camouflaged detection, polyp segmentation, transparent object detection, and defect inspection, objects blend with surroundings through physical processes that anti-correlate changes in illumination and reflectance. Retinex theory allows homogeneous decomposition of an image into these two components while staying in the original spatial domain, unlike frequency-based methods. Because the components are not forced to match simultaneously, the decomposition preserves or increases total foreground-background discriminability, with the gain largest when the anti-correlation is strongest. The authors formalize this as the Discriminability Gap Theorem and build RIDE around a learned, task-driven Retinex module, an attention mechanism that selects helpful regions, and a contrastive loss in reflectance space.

Core claim

Across diverse COS sub-tasks the underlying physical processes systematically anti-correlate illumination and reflectance differences, yielding theoretical guarantees that Retinex decomposition preserves or strictly improves total foreground-background discriminability across the full physical regime, with anti-correlation maximizing the gain. The Discriminability Gap Theorem states that visual entanglement enforces appearance matching in the composite image but does not require simultaneous matching in both component spaces.

What carries the argument

The Discriminability Gap Theorem together with the Task-Driven Retinex Decomposition module that learns segmentation-optimal factorizations end-to-end.

If this is right

  • Retinex decomposition can be inserted into existing COS pipelines to improve pixel-aligned cues without redistributing evidence across scales or frequencies.
  • The gain from decomposition is maximized when the physical anti-correlation between illumination and reflectance is strongest.
  • A contrastive loss applied in reflectance feature space helps break appearance matching that survives in the original RGB image.
  • Adaptive attention can selectively apply the decomposition only where it increases discriminability.

Where Pith is reading between the lines

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

  • The same anti-correlation principle may apply to other dense-prediction problems that involve lighting or surface variation, such as low-light enhancement or medical imaging under varying illumination.
  • If the theorem holds, it supplies a principled reason to prefer homogeneous over heterogeneous decompositions when spatial alignment of cues matters.
  • One could test whether learned Retinex modules transfer across COS sub-tasks without retraining the decomposition network.

Load-bearing premise

Physical processes in COS tasks systematically anti-correlate illumination and reflectance differences.

What would settle it

A dataset of COS images in which measured illumination and reflectance differences are positively correlated or uncorrelated and where applying Retinex decomposition produces no gain or a loss in foreground-background separation.

Figures

Figures reproduced from arXiv: 2605.15450 by Chengyu Fang, Chunming He, Dingming Zhang, Fengyang Xiao, Jingjia Feng, Longxiang Tang, Rihan Zhang, Sina Farsiu.

Figure 1
Figure 1. Figure 1: Heterogeneous vs. homogeneous decomposition for COS. Heterogeneous methods (left) transform images to different representation spaces (frequency domain), scattering spatial information. Our homogeneous approach (right) decomposes within the same spatial domain via Retinex theory, preserving locality while exposing concealed material differences. A natural strategy to handle such visual entanglement is imag… view at source ↗
Figure 2
Figure 2. Figure 2: Overall architecture of RIDE. where c indexes color channels. ∇ = (∇h, ∇v) denotes spatial gradients along horizontal and vertical directions. LME represents Mutual Exclusivity (ME) loss, which encourages edges to be attributed to either illumination or reflectance, but not both: LME = 1 |Ω| X p∈Ω X d∈{h,v} |∇dL(p)| · |∇dR(p)|. (5) This is crucial for COS: clean reflectance edges indicate material (and pot… view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative comparison. We also present learned Retinex decomposition ( [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: In-depth analysis. (a) ∆Sα vs. ρ bins, validating Theorem 1. (b) Dominant component across COS sub-tasks. (c) Task-driven decomposition yields larger gaps than fixed Retinex. Alternative homogeneous decompositions. As indicated in [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
read the original abstract

Concealed Object Segmentation (COS) encompasses a family of dense-prediction tasks, including camouflaged object detection, polyp segmentation, transparent object detection, and industrial defect inspection, where targets are visually entangled with their surroundings through different physical mechanisms. Existing methods either operate directly on RGB images or employ \emph{heterogeneous} decompositions (\eg, Fourier, wavelet) that redistribute spatial evidence across scale/frequency coefficients, making pixel-aligned cues less direct. We introduce a fundamentally different perspective: \textbf{homogeneous image decomposition} via Retinex theory, which factorizes an image into illumination and reflectance components within the \emph{same} spatial domain. Our key insight is that visual entanglement enforces appearance matching in the composite space, but this does \emph{not} necessitate simultaneous matching in both component spaces, a phenomenon we formalize as the \textbf{Discriminability Gap Theorem}. Crucially, we show that across diverse COS sub-tasks, the underlying physical processes systematically anti-correlate illumination and reflectance differences, yielding theoretical guarantees that Retinex decomposition preserves or strictly improves total foreground--background discriminability across the full physical regime, with anti-correlation maximizing the gain. Building on this, we propose \textbf{RIDE} comprising: (i) a Task-Driven Retinex Decomposition module that learns segmentation-optimal factorizations end-to-end; (ii) a Discriminability Gap Attention mechanism that adaptively exploits where decomposition helps; and (iii) a Camouflage-Breaking Contrastive loss operating in reflectance feature space.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 3 minor

Summary. The manuscript introduces RIDE, a framework for Concealed Object Segmentation (COS) tasks including camouflaged object detection, polyp segmentation, transparent object detection, and industrial defect inspection. It advocates homogeneous Retinex decomposition to factorize images into illumination and reflectance components in the same spatial domain, contrasting with heterogeneous decompositions. The central theoretical contribution is the Discriminability Gap Theorem, which asserts that physical processes in COS sub-tasks systematically anti-correlate illumination and reflectance differences. This is claimed to yield guarantees that Retinex decomposition preserves or strictly improves total foreground-background discriminability across the full physical regime, with anti-correlation maximizing the gain. The method comprises a Task-Driven Retinex Decomposition module, Discriminability Gap Attention, and a Camouflage-Breaking Contrastive loss in reflectance feature space.

Significance. If the anti-correlation assumption and Discriminability Gap Theorem are rigorously established with supporting derivations or measurements, the work could provide a principled alternative to existing COS methods by offering theoretical guarantees on improved discriminability via homogeneous decomposition. The end-to-end trainable decomposition and contrastive loss in reflectance space represent practical strengths that could translate to better performance on visually entangled targets. The approach distinguishes itself by focusing on same-domain factorization rather than scale/frequency redistribution.

major comments (2)
  1. [Discriminability Gap Theorem] Discriminability Gap Theorem (as stated in the abstract and formalized in the theoretical section): The claim that underlying physical processes 'systematically anti-correlate illumination and reflectance differences' across camouflage, polyp, transparent-object, and defect mechanisms is presented as an empirical fact supporting the guarantees. However, the manuscript provides neither a first-principles physical model deriving the sign of the correlation for each mechanism nor quantitative verification (e.g., measured correlation coefficients or statistical tests on real COS images). This assumption is load-bearing for the theorem's guarantees; if the anti-correlation is not uniformly negative or is condition-dependent, the claimed preservation or strict improvement in discriminability does not follow.
  2. [§3] §3 (theoretical development) and experimental validation: The theorem is invoked to guarantee improvement 'across the full physical regime,' yet the paper does not report ablation studies isolating the contribution of the anti-correlation property versus the learned decomposition or contrastive loss. If empirical gains are observed, they may be attributable to the network components rather than the asserted theoretical mechanism, weakening the central claim that the decomposition 'preserves or strictly improves' discriminability by construction.
minor comments (3)
  1. [Notation and definitions] Clarify the precise mathematical definition of 'total foreground--background discriminability' (including any distance metric or separability measure used in the theorem) to allow independent verification.
  2. [Method] In the method description, explicitly state how the Task-Driven Retinex Decomposition module is optimized end-to-end (e.g., which loss terms supervise the illumination and reflectance branches) to improve reproducibility.
  3. [Discussion] Add a brief discussion of potential failure cases where illumination-reflectance anti-correlation may not hold (e.g., under specific lighting or material conditions) to temper the 'full physical regime' claim.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback. We address each major comment point by point below, providing clarifications on the theoretical claims and committing to revisions that strengthen the empirical support.

read point-by-point responses

  1. Referee: [Discriminability Gap Theorem] Discriminability Gap Theorem (as stated in the abstract and formalized in the theoretical section): The claim that underlying physical processes 'systematically anti-correlate illumination and reflectance differences' across camouflage, polyp, transparent-object, and defect mechanisms is presented as an empirical fact supporting the guarantees. However, the manuscript provides neither a first-principles physical model deriving the sign of the correlation for each mechanism nor quantitative verification (e.g., measured correlation coefficients or statistical tests on real COS images). This assumption is load-bearing for the theorem's guarantees; if the anti-correlation is not uniformly negative or is condition-dependent, the claimed preservation or strict improvement in discriminability does not follow.

    Authors: The Discriminability Gap Theorem is a conditional mathematical statement: whenever illumination and reflectance differences are anti-correlated, the homogeneous decomposition is guaranteed to preserve or strictly increase total foreground-background discriminability. The manuscript motivates the anti-correlation through task-specific physical reasoning (camouflage via reflectance matching compensated by 3D illumination variation; polyps via specular illumination mismatch on similar reflectance; transparent objects via refraction-induced illumination differences; defects via surface-normal effects). While a single first-principles derivation spanning all mechanisms is not derived, we support the assumption with per-task analysis and correlation measurements in Section 4.3 and the supplement. We will add explicit tables of Pearson correlation coefficients with statistical tests across all datasets in the revision to quantify the sign and strength of the anti-correlation. revision: yes

  2. Referee: [§3] §3 (theoretical development) and experimental validation: The theorem is invoked to guarantee improvement 'across the full physical regime,' yet the paper does not report ablation studies isolating the contribution of the anti-correlation property versus the learned decomposition or contrastive loss. If empirical gains are observed, they may be attributable to the network components rather than the asserted theoretical mechanism, weakening the central claim that the decomposition 'preserves or strictly improves' discriminability by construction.

    Authors: We agree that isolating the anti-correlation mechanism is necessary to substantiate the theoretical contribution. Existing ablations demonstrate the value of the Task-Driven Retinex Decomposition and Camouflage-Breaking Contrastive loss. To directly test the anti-correlation, we will add a controlled ablation in the revision: we introduce a positive-correlation regularization term during decomposition training and compare performance against the standard (anti-correlated) setting. A performance drop under positive correlation would support that gains derive from the theorem's mechanism rather than architecture alone. These results will be reported with statistical significance. revision: yes

Circularity Check

0 steps flagged

No significant circularity; theorem formalizes independent physical observation

full rationale

The paper introduces the Discriminability Gap Theorem by formalizing that visual entanglement enforces matching in composite RGB space but not necessarily in separate illumination and reflectance components. It then states that physical processes across COS sub-tasks systematically anti-correlate illumination and reflectance differences, which yields guarantees that Retinex decomposition preserves or improves foreground-background discriminability. This chain does not reduce any prediction or result to its inputs by construction, nor does it rely on self-citation chains, fitted parameters renamed as predictions, or smuggled ansatzes. The anti-correlation is presented as an observed property of the physical mechanisms that supports the theorem, and the subsequent RIDE modules (task-driven decomposition, attention, contrastive loss) are built on top without the central claim collapsing into a tautology. The derivation is therefore self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on Retinex theory as a domain assumption and on the unproven (in abstract) Discriminability Gap Theorem; no free parameters or invented entities are mentioned.

axioms (1)
  • domain assumption Retinex theory factorizes an image into illumination and reflectance components within the same spatial domain.
    Invoked in the abstract as the basis for homogeneous decomposition that keeps pixel-aligned cues direct.

pith-pipeline@v0.9.0 · 5834 in / 1229 out tokens · 82960 ms · 2026-05-19T15:04:51.135807+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

73 extracted references · 73 canonical work pages · 2 internal anchors

  1. [1]

    Camouflaged object detection

    Deng-Ping Fan, Ge-Peng Ji, Guolei Sun, Ming-Ming Cheng, and Jianbing Shen. Camouflaged object detection. InCVPR, pages 2777–2787, 2020. 1, 2

    work page 2020

  2. [2]

    Anabranch network for camouflaged object segmentation.Comput

    Trung-Nghia Le, Tam V Nguyen, Zhongliang Nie, Minh-Triet Tran, and Akihiro Sugimoto. Anabranch network for camouflaged object segmentation.Comput. Vis. Image Underst., 184:45– 56, 2019. 1, 2

    work page 2019

  3. [3]

    Pranet: Parallel reverse attention network for polyp segmentation

    Deng-Ping Fan, Ge-Peng Ji, and Tao Zhou. Pranet: Parallel reverse attention network for polyp segmentation. InMICCAI, pages 263–273, 2020. 1, 2, 3

    work page 2020

  4. [4]

    Scaler: Sam-enhanced collaborative learning for label-deficient concealed object segmentation.arXiv preprint arXiv:2511.18136, 2025

    Chunming He, Rihan Zhang, Longxiang Tang, Ziyun Yang, Kai Li, Deng-Ping Fan, and Sina Farsiu. Scaler: Sam-enhanced collaborative learning for label-deficient concealed object segmentation.arXiv preprint arXiv:2511.18136, 2025. 1

    work page arXiv 2025

  5. [5]

    Refining context-entangled content segmentation via curriculum selection and anti-curriculum promotion.arXiv preprint arXiv:2602.01183, 2026

    Chunming He, Rihan Zhang, Fengyang Xiao, Dingming Zhang, Zhiwen Cao, and Sina Farsiu. Refining context-entangled content segmentation via curriculum selection and anti-curriculum promotion.arXiv preprint arXiv:2602.01183, 2026. 1

    work page arXiv 2026

  6. [6]

    Concealed object detection

    Deng-Ping Fan, Ge-Peng Ji, Ming-Ming Cheng, and Ling Shao. Concealed object detection. IEEE Trans. Pattern Anal. Mach. Intell., 2021. 1

    work page 2021

  7. [7]

    Mutual graph learning for camouflaged object detection

    Qiang Zhai, Xin Li, Fan Yang, and Chenglizhao Chen. Mutual graph learning for camouflaged object detection. InCVPR, pages 2997–3007, 2021. 1

    work page 2021

  8. [8]

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

    Hongwei Zhu, Peng Li, Haoran Xie, Xuefeng Yan, Dong Liang, Dapeng Chen, Mingqiang Wei, and Jing Qin. I can find you! boundary-guided separated attention network for camouflaged object detection. InAAAI, volume 36, pages 3608–3616, 2022. 1, 7

    work page 2022

  9. [9]

    Detecting camouflaged object in frequency domain

    Yijie Zhong, Bo Li, Lv Tang, and Senyun Kuang. Detecting camouflaged object in frequency domain. InCVPR, pages 4504–4513, 2022. 2, 3

    work page 2022

  10. [10]

    Camouflaged object detection with feature decomposition and edge reconstruction

    Chunming He, Kai Li, Yachao Zhang, Longxiang Tang, and Yulun Zhang. Camouflaged object detection with feature decomposition and edge reconstruction. InCVPR, pages 22046–22055,

    work page

  11. [11]

    Fre- quency perception network for camouflaged object detection

    Runmin Cong, Mengyao Sun, Sanyi Zhang, Xiaofei Zhou, Wei Zhang, and Yao Zhao. Fre- quency perception network for camouflaged object detection. InProceedings of the 31st ACM international conference on multimedia, pages 1179–1189, 2023. 2

    work page 2023

  12. [12]

    Lightness and retinex theory.Journal of the Optical society of America, 61(1):1–11, 1971

    Edwin H Land and John J McCann. Lightness and retinex theory.Journal of the Optical society of America, 61(1):1–11, 1971. 2, 3

    work page 1971

  13. [13]

    Animal camouflage: current issues and new perspectives

    Martin Stevens and Sami Merilaita. Animal camouflage: current issues and new perspectives. Philosophical Transactions of the Royal Society B: Biological Sciences, 364(1516):423–427,

    work page

  14. [14]

    A survey of camouflaged object detection and beyond.arXiv preprint arXiv:2408.14562, 2024

    Fengyang Xiao, Sujie Hu, Yuqi Shen, and Chunming He. A survey of camouflaged object detection and beyond.arXiv preprint arXiv:2408.14562, 2024. 2

    work page arXiv 2024

  15. [15]

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

    Haiyang Mei, Xin Yang, Yang Wang, Yuanyuan Liu, and Shengfeng He. Don’t hit me! glass detection in real-world scenes. InCVPR, pages 3687–3696, 2020. 2, 3, 8

    work page 2020

  16. [16]

    Segment concealed object with incomplete supervision.IEEE Trans

    Chunming He, Kai Li, Yachao Zhang, Ziyun Yang, Longxiang Tang, Yulun Zhang, Linghe Kong, and Sina Farsiu. Segment concealed object with incomplete supervision.IEEE Trans. Pattern Anal. Mach. Intell., 2025. 3

    work page 2025

  17. [17]

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

    Youwei Pang, Xiaoqi Zhao, Tian-Zhu Xiang, Lihe Zhang, and Huchuan Lu. Zoom in and out: A mixed-scale triplet network for camouflaged object detection. InCVPR, pages 2160–2170,

    work page

  18. [18]

    Segment, magnify and reiterate: Detect camouflaged objects hard way

    Qi Jia, Shuilian Yao, and Yu Liu. Segment, magnify and reiterate: Detect camouflaged objects hard way. InCVPR, pages 713–722, 2022. 3 11

    work page 2022

  19. [19]

    Sam-cod: Sam-guided unified framework for weakly-supervised camouflaged object detection.arXiv preprint arXiv:2408.10760, 2024

    Huafeng Chen, Pengxu Wei, Guangqian Guo, and Shan Gao. Sam-cod: Sam-guided unified framework for weakly-supervised camouflaged object detection.arXiv preprint arXiv:2408.10760, 2024. 3

    work page arXiv 2024

  20. [20]

    Reversible unfolding network for concealed visual perception with generative refinement.arXiv preprint arXiv:2508.15027, 2025

    Chunming He, Fengyang Xiao, Rihan Zhang, Chengyu Fang, Deng-Ping Fan, and Sina Farsiu. Reversible unfolding network for concealed visual perception with generative refinement.arXiv preprint arXiv:2508.15027, 2025. 3

    work page arXiv 2025

  21. [21]

    Uncertainty-masked bernoulli diffusion for camouflaged object detection refinement.arXiv preprint arXiv:2506.10712, 2025

    Yuqi Shen, Fengyang Xiao, Sujie Hu, Youwei Pang, Yifan Pu, Chengyu Fang, Xiu Li, and Chunming He. Uncertainty-masked bernoulli diffusion for camouflaged object detection refinement.arXiv preprint arXiv:2506.10712, 2025. 3

    work page arXiv 2025

  22. [22]

    Enhanced boundary learning for glass-like object segmentation

    Hao He, Xiangtai Li, Guangliang Cheng, Jianping Shi, Yunhai Tong, Gaofeng Meng, Véronique Prinet, and LuBin Weng. Enhanced boundary learning for glass-like object segmentation. In ICCV, pages 15859–15868, 2021. 3, 8

    work page 2021

  23. [23]

    Deep Retinex Decomposition for Low-Light Enhancement

    Chen Wei, Wenjing Wang, Wenhan Yang, and Jiaying Liu. Deep retinex decomposition for low-light enhancement.arXiv preprint arXiv:1808.04560, 2018. 3, 9

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  24. [24]

    Uretinex- net: Retinex-based deep unfolding network for low-light image enhancement

    Wenhui Wu, Jian Weng, Pingping Zhang, Xu Wang, Wenhan Yang, and Jianmin Jiang. Uretinex- net: Retinex-based deep unfolding network for low-light image enhancement. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 5901–5910,

    work page

  25. [25]

    Reti-diff: Illumination degradation image restoration with retinex-based latent diffusion model.ICLR, 2025

    Chunming He, Chengyu Fang, Yulun Zhang, Tian Ye, Kai Li, Longxiang Tang, Zhenhua Guo, Xiu Li, and Sina Farsiu. Reti-diff: Illumination degradation image restoration with retinex-based latent diffusion model.ICLR, 2025. 3

    work page 2025

  26. [26]

    Unfoldir: Rethinking deep unfolding network in illumination degradation image restoration.arXiv preprint arXiv:2505.06683, 2025

    Chunming He, Rihan Zhang, Fengyang Xiao, Chengyu Fang, Longxiang Tang, Yulun Zhang, and Sina Farsiu. Unfoldir: Rethinking deep unfolding network in illumination degradation image restoration.arXiv preprint arXiv:2505.06683, 2025. 3

    work page arXiv 2025

  27. [27]

    UnfoldLDM: Degradation-Aware Unfolding with Iterative Latent Diffusion Priors for Blind Image Restoration

    Chunming He, Rihan Zhang, Zheng Chen, Bowen Yang, CHengyu Fang, Yunlong Lin, Fengyang Xiao, and Sina Farsiu. Unfoldldm: Deep unfolding-based blind image restoration with latent diffusion priors.arXiv preprint arXiv:2511.18152, 2025. 3

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  28. [28]

    Recovering intrinsic scene characteris- tics.Comput

    Harry Barrow, J Tenenbaum, A Hanson, and E Riseman. Recovering intrinsic scene characteris- tics.Comput. vis. syst, 2(3-26):2, 1978. 3

    work page 1978

  29. [29]

    Beyond ground-truth: Leveraging image quality priors for real-world image restoration.CVPR, 2026

    Fengyang Xiao, Peng Hu, Lei Xu, Xinge Guo, Guanyi Qin, Yuqi Shen, Chengyu Fang, Rihan Zhang, Chunming He, and Sina Farsiu. Beyond ground-truth: Leveraging image quality priors for real-world image restoration.CVPR, 2026. 3

    work page 2026

  30. [30]

    Structure extraction from texture via relative total variation.ACM transactions on graphics (TOG), 31(6):1–10, 2012

    Li Xu, Qiong Yan, Yang Xia, and Jiaya Jia. Structure extraction from texture via relative total variation.ACM transactions on graphics (TOG), 31(6):1–10, 2012. 3

    work page 2012

  31. [31]

    Qualiteacher: Quality-conditioned pseudo-labeling for real-world image restoration.arXiv preprint arXiv:2603.08030, 2026

    Fengyang Xiao, Jingjia Feng, Peng Hu, Dingming Zhang, Lei Xu, Guanyi Qin, Lu Li, Chunming He, and Sina Farsiu. Qualiteacher: Quality-conditioned pseudo-labeling for real-world image restoration.arXiv preprint arXiv:2603.08030, 2026. 3

    work page arXiv 2026

  32. [32]

    Guided image filtering.IEEE transactions on pattern analysis and machine intelligence, 35(6):1397–1409, 2012

    Kaiming He, Jian Sun, and Xiaoou Tang. Guided image filtering.IEEE transactions on pattern analysis and machine intelligence, 35(6):1397–1409, 2012. 3

    work page 2012

  33. [33]

    Hqg-net: Unpaired medical image enhancement with high-quality guidance.IEEE Trans

    Guoxia Xu, Jiangpeng Yan, Longxiang Tang, and Yulun Zhang. Hqg-net: Unpaired medical image enhancement with high-quality guidance.IEEE Trans. Neural Networks Learn. Syst.,

    work page

  34. [34]

    Disentangling task-oriented representations for unsupervised domain adaptation.IEEE Transactions on Image Processing, 31:1012–1026, 2021

    Pingyang Dai, Peixian Chen, Qiong Wu, Xiaopeng Hong, Qixiang Ye, Qi Tian, Chia-Wen Lin, and Rongrong Ji. Disentangling task-oriented representations for unsupervised domain adaptation.IEEE Transactions on Image Processing, 31:1012–1026, 2021. 3

    work page 2021

  35. [35]

    Learning in the frequency domain

    Kai Xu, Minghai Qin, Fei Sun, Yuhao Wang, Yen-Kuang Chen, and Fengbo Ren. Learning in the frequency domain. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 1740–1749, 2020. 3 12

    work page 2020

  36. [36]

    Global filter networks for image classification.Advances in neural information processing systems, 34:980–993, 2021

    Yongming Rao, Wenliang Zhao, Zheng Zhu, Jiwen Lu, and Jie Zhou. Global filter networks for image classification.Advances in neural information processing systems, 34:980–993, 2021. 3

    work page 2021

  37. [37]

    Fast fourier convolution.Advances in Neural Information Processing Systems, 33:4479–4488, 2020

    Lu Chi, Borui Jiang, and Yadong Mu. Fast fourier convolution.Advances in Neural Information Processing Systems, 33:4479–4488, 2020. 3

    work page 2020

  38. [38]

    Independent component analysis: algorithms and applications

    Aapo Hyvärinen and Erkki Oja. Independent component analysis: algorithms and applications. Neural networks, 13(4-5):411–430, 2000. 3

    work page 2000

  39. [39]

    Trace ratio vs

    Huan Wang, Shuicheng Yan, Dong Xu, Xiaoou Tang, and Thomas Huang. Trace ratio vs. ratio trace for dimensionality reduction. In2007 IEEE Conference on Computer Vision and Pattern Recognition, pages 1–8. IEEE, 2007. 3

    work page 2007

  40. [40]

    Sparse trace ratio lda for supervised feature selection.IEEE transactions on cybernetics, 54(4):2420–2433, 2023

    Zhengxin Li, Feiping Nie, Danyang Wu, Zheng Wang, and Xuelong Li. Sparse trace ratio lda for supervised feature selection.IEEE transactions on cybernetics, 54(4):2420–2433, 2023. 3

    work page 2023

  41. [41]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. InMICCAI, pages 234–241. Springer, 2015. 4

    work page 2015

  42. [42]

    Run: Reversible unfolding network for concealed object segmentation.ICML, 2025

    Chunming He, Rihan Zhang, Fengyang Xiao, Chenyu Fang, Longxiang Tang, Yulun Zhang, Linghe Kong, Deng-Ping Fan, Kai Li, and Sina Farsiu. Run: Reversible unfolding network for concealed object segmentation.ICML, 2025. 6, 7, 8

    work page 2025

  43. [43]

    Imagenet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, and Kai Li. Imagenet: A large-scale hierarchical image database. InCVPR, pages 248–255, 2009. 6

    work page 2009

  44. [44]

    Exploring figure-ground assignment mechanism in perceptual organization

    Wei Zhai, Yang Cao, and Jing Zhang. Exploring figure-ground assignment mechanism in perceptual organization. InNeurIPS, volume 35, 2023. 7

    work page 2023

  45. [45]

    Focusd- iffuser: Perceiving local disparities for camouflaged object detection

    Jianwei Zhao, Xin Li, Fan Yang, Qiang Zhai, Ao Luo, Zicheng Jiao, and Hong Cheng. Focusd- iffuser: Perceiving local disparities for camouflaged object detection. InECCV, pages 181–198,

    work page

  46. [46]

    Frequency-spatial entangle- ment learning for camouflaged object detection

    Yanguang Sun, Chunyan Xu, Jian Yang, Hanyu Xuan, and Lei Luo. Frequency-spatial entangle- ment learning for camouflaged object detection. InECCV, pages 343–360, 2024. 7

    work page 2024

  47. [47]

    Conditional diffusion models for camouflaged and salient object detection.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

    Ke Sun, Zhongxi Chen, Xianming Lin, Xiaoshuai Sun, Hong Liu, and Rongrong Ji. Conditional diffusion models for camouflaged and salient object detection.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025. 7

    work page 2025

  48. [48]

    Vscode: General visual salient and camouflaged object detection with 2d prompt learning

    Ziyang Luo, Nian Liu, Wangbo Zhao, Xuguang Yang, Dingwen Zhang, Deng-Ping Fan, Fahad Khan, and Junwei Han. Vscode: General visual salient and camouflaged object detection with 2d prompt learning. InCVPR, pages 17169–17180, 2024. 7

    work page 2024

  49. [49]

    C3net: Context-contrast network for camouflaged object detection.arXiv preprint arXiv:2511.12627,

    Baber Jan, Aiman H El-Maleh, Abdul Jabbar Siddiqui, Abdul Bais, and Saeed Anwar. C3net: Context-contrast network for camouflaged object detection.arXiv preprint arXiv:2511.12627,

    work page arXiv

  50. [50]

    Camofocus: Enhancing camouflage object detection with split-feature focal modulation and context refinement

    Abbas Khan, Mustaqeem Khan, Wail Gueaieb, Abdulmotaleb El Saddik, Giulia De Masi, and Fakhri Karray. Camofocus: Enhancing camouflage object detection with split-feature focal modulation and context refinement. InWACV, pages 1434–1443, 2024. 7

    work page 2024

  51. [51]

    Toward embedded detection of polyps in wce images for early diagnosis.Int

    Juan Silva, Aymeric Histace, Olivier Romain, and Xavier Dray. Toward embedded detection of polyps in wce images for early diagnosis.Int. J. Comput. Assist. Radiol. Surg., 9:283–293,

    work page

  52. [52]

    Polyp-pvt: Polyp segmentation with pyramid vision transformers.CAAI Artif

    Bo Dong, Wenhai Wang, Deng-Ping Fan, Jinpeng Li, Huazhu Fu, and Ling Shao. Polyp-pvt: Polyp segmentation with pyramid vision transformers.CAAI Artif. Intell. Res., 2, 2023. 8

    work page 2023

  53. [53]

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

    Samir Jain, Rohan Atale, Anubhav Gupta, Utkarsh Mishra, Ayan Seal, Aparajita Ojha, Joanna Jaworek-Korjakowska, and Ondrej Krejcar. Coinnet: A convolution-involution network with a novel statistical attention for automatic polyp segmentation.IEEE Trans. Med. Imaging, 42(12):3987–4000, 2023. 8 13

    work page 2023

  54. [54]

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

    Wei Wang, Huiying Sun, and Xin Wang. Lssnet: A method for colon polyp segmentation based on local feature supplementation and shallow feature supplementation. InMICCAI, pages 446–456. Springer, 2024. 8

    work page 2024

  55. [55]

    Meganet: Multi-scale edge-guided attention network for weak boundary polyp segmentation

    Nhat-Tan Bui, Dinh-Hieu Hoang, Quang-Thuc Nguyen, Minh-Triet Tran, and Ngan Le. Meganet: Multi-scale edge-guided attention network for weak boundary polyp segmentation. InWACV, pages 7985–7994, 2024. 8

    work page 2024

  56. [56]

    Rfenet: towards reciprocal feature evolution for glass segmentation

    Ke Fan, Changan Wang, Yabiao Wang, Chengjie Wang, Ran Yi, and Lizhuang Ma. Rfenet: towards reciprocal feature evolution for glass segmentation. InIJCAI, pages 717–725, 2023. 8

    work page 2023

  57. [57]

    Internal-external boundary attention fusion for glass surface segmentation.arXiv preprint arXiv:2307.00212, 2024

    Dongshen Han, Seungkyu Lee, Chaoning Zhang, Heechan Yoon, Hyukmin Kwon, Hyun- Cheol Kim, and Hyon-Gon Choo. Internal-external boundary attention fusion for glass surface segmentation.arXiv preprint arXiv:2307.00212, 2024. 8

    work page arXiv 2024

  58. [58]

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

    Tao Yan, Jiahui Gao, Ke Xu, Xiangjie Zhu, Hao Huang, Helong Li, Benjamin Wah, and Rynson WH Lau. Ghostingnet: A novel approach for glass surface detection with ghosting cues. IEEE Trans. Pattern Anal. Mach. Intell., 2024. 8

    work page 2024

  59. [59]

    Advances in deep concealed scene understanding.Visual Intell., 1(1):16, 2023

    Deng-Ping Fan, Ge-Peng Ji, Peng Xu, Ming-Ming Cheng, Christos Sakaridis, and Luc Van Gool. Advances in deep concealed scene understanding.Visual Intell., 1(1):16, 2023. 8

    work page 2023

  60. [60]

    High-resolution iterative feedback network for camouflaged object detection

    Xiaobin Hu, Shuo Wang, Xuebin Qin, Hang Dai, Wenqi Ren, Donghao Luo, Ying Tai, and Ling Shao. High-resolution iterative feedback network for camouflaged object detection. InAAAI, volume 37, pages 881–889, 2023. 8

    work page 2023

  61. [61]

    Camoformer: Masked separable attention for camouflaged object detection

    Bowen Yin, Xuying Zhang, Deng-Ping Fan, Shaohui Jiao, Ming-Ming Cheng, Luc Van Gool, and Qibin Hou. Camoformer: Masked separable attention for camouflaged object detection. IEEE Trans. Pattern Anal. Mach. Intell., 2024. 8

    work page 2024

  62. [62]

    Oaformer: Occlusion aware trans- former for camouflaged object detection

    Xin Yang, Hengliang Zhu, Guojun Mao, and Shuli Xing. Oaformer: Occlusion aware trans- former for camouflaged object detection. InICME, pages 1421–1426. IEEE, 2023. 8

    work page 2023

  63. [63]

    Scene parsing through ade20k dataset

    Bolei Zhou, Hang Zhao, Xavier Puig, Sanja Fidler, Adela Barriuso, and Antonio Torralba. Scene parsing through ade20k dataset. InCVPR, pages 633–641, 2017. 8

    work page 2017

  64. [64]

    Pem: Prototype-based efficient maskformer for im- age segmentation

    Niccolo Cavagnero, Gabriele Rosi, Claudia Cuttano, Francesca Pistilli, Marco Ciccone, Giuseppe Averta, and Fabio Cermelli. Pem: Prototype-based efficient maskformer for im- age segmentation. InCVPR, pages 15804–15813, 2024. 8

    work page 2024

  65. [65]

    Microsoft coco: Common objects in context

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. Microsoft coco: Common objects in context. InEuropean conference on computer vision, pages 740–755. Springer, 2014. 8

    work page 2014

  66. [66]

    Masked-attention mask transformer for universal image segmentation

    Bowen Cheng, Ishan Misra, Alexander G Schwing, Alexander Kirillov, and Rohit Girdhar. Masked-attention mask transformer for universal image segmentation. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 1290–1299, 2022. 8

    work page 2022

  67. [67]

    Large-scale training of shadow detectors with noisily-annotated shadow examples

    Tomás F Yago Vicente, Le Hou, Chen-Ping Yu, Minh Hoai, and Dimitris Samaras. Large-scale training of shadow detectors with noisily-annotated shadow examples. InEuropean conference on computer vision, pages 816–832. Springer, 2016. 8

    work page 2016

  68. [68]

    Spider: a unified framework for context-dependent concept segmentation

    Xiaoqi Zhao, Youwei Pang, Wei Ji, Baicheng Sheng, Jiaming Zuo, Lihe Zhang, and Huchuan Lu. Spider: a unified framework for context-dependent concept segmentation. InProceedings of the 41st International Conference on Machine Learning, pages 60906–60926, 2024. 8

    work page 2024

  69. [69]

    Isnet: Shape matters for infrared small target detection

    Mingjin Zhang, Rui Zhang, Yuxiang Yang, Haichen Bai, Jing Zhang, and Jie Guo. Isnet: Shape matters for infrared small target detection. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 877–886, 2022. 8

    work page 2022

  70. [70]

    Irsam: Advancing segment anything model for infrared small target detection

    Mingjin Zhang, Yuchun Wang, Jie Guo, Yunsong Li, Xinbo Gao, and Jing Zhang. Irsam: Advancing segment anything model for infrared small target detection. InEuropean Conference on Computer Vision, pages 233–249. Springer, 2024. 8 14

    work page 2024

  71. [71]

    Learning to detect salient objects with image-level supervision

    Lijun Wang, Huchuan Lu, and Yifan Wang. Learning to detect salient objects with image-level supervision. InCVPR, pages 136–145, 2017. 8

    work page 2017

  72. [72]

    Implicit motion handling for video camouflaged object detection

    Xuelian Cheng, Huan Xiong, Deng-Ping Fan, Yiran Zhong, Mehrtash Harandi, Tom Drummond, and Zongyuan Ge. Implicit motion handling for video camouflaged object detection. InCVPR, pages 13864–13873, 2022. 8

    work page 2022

  73. [73]

    Zoomnext: A unified collaborative pyramid network for camouflaged object detection.IEEE Trans

    Youwei Pang, Xiaoqi Zhao, Tian-Zhu Xiang, Lihe Zhang, and Huchuan Lu. Zoomnext: A unified collaborative pyramid network for camouflaged object detection.IEEE Trans. Pattern Anal. Mach. Intell., 2024. 8 15

    work page 2024