Pith. sign in

REVIEW 4 major objections 5 minor 2 cited by

Uncertainty-Masked Bernoulli Diffusion for Camouflaged Object Detection Refinement

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read UMBD, a plug-in uncertainty-masked Bernoulli diffusion refiner, claims to improve coarse masks from existing camouflaged-object detectors by 5.5% in MAE and 3.2% in weighted F-measure on four COD benchmarks.

desk verdict Empirically useful plug-in COD refiner, but Eq. (1) has a real mathematical hole that undercuts the stated residual-uncertainty mechanism; deserve peer review with major revisions. read the letter →

arxiv 2506.10712 v1 pith:4CHY7GZU submitted 2025-06-12 cs.CV

Yuqi Shen , Fengyang Xiao , Sujie Hu , Youwei Pang , Yifan Pu , Chengyu Fang , Xiu Li , Chunming He This is my paper
classification cs.CV
keywords camouflagedobjectdetectionsegmentationrefinementBernoullidiffusionuncertaintyestimationgenerativeresiduallearningplug-and-playrefinerconcealed
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 UMBD, a post-processing refiner for camouflaged object detection that works on top of any existing encoder-decoder detector. Its central claim is that by running a Bernoulli diffusion process only on the residual region where a coarse mask is uncertain, instead of on the whole mask, the refiner consistently improves coarse predictions. Across CHAMELEON, CAMO, COD10K, and NC4K, UMBD reports average gains of 5.5% in mean absolute error and 3.2% in weighted F-measure, with gains visible for CNN and transformer detectors and with refinement costs kept low by a three- to ten-step sampling schedule. If true, UMBD gives a generic way to squeeze extra accuracy out of existing COD models without retraining them, and the same recipe extends to polyp segmentation and transparent object detection.

What carries the argument

The central object is the residual refinement identity $M_r = U \odot M_{GT} + (1-U) \odot M_c$, where $U$ is a pixelwise uncertainty map. The mechanism that carries the argument is an uncertainty-masked Bernoulli diffusion process: the forward process interpolates between the masked ground truth and the masked coarse mask with Bernoulli noise, and the reverse process, initialized from $\mathrm{B}(M_c)$, denoises only the uncertain residual. A second load-bearing component is HUQNet, a two-branch uncertainty estimator that fuses a Bayesian neural network's variance map, an entropy map, and a discriminative decoder through window-based cross-attention to produce $\hat U = h(x,M_c)$; feature adaptation fuses the frozen detector's encoder features into the U-Net denoiser, and DDIM acceleration allows inference in as few as three steps.

What would settle it

Run UMBD inference twice on COD10K: once with the learned $\hat U$ and once with the oracle $U_{GT}=|M_c-M_{GT}|$ in place of it. If the oracle version does not clearly beat the learned version, the masking mechanism rather than HUQNet is carrying the gain; if oracle gains far exceed the reported 5.5% MAE improvement, then the method's central promise depends on closing the uncertainty-estimation gap. A second check: feed a fixed random map with the same overall magnitude as $\hat U$; if random masking still improves coarse masks, the targeted uncertainty guidance is not essential.

Watch

Extended reading notes

Core claim

The paper's discovery claim is that camouflaged-object mask refinement can be recast as residual generation under an uncertainty mask. Writing $M_c$ for a coarse mask and $M_{GT}$ for ground truth, the ideal refined mask satisfies $M_r = U \odot M_{GT} + (1-U) \odot M_c$ with $U=|M_c-M_{GT}|$, so only wrongly segmented pixels need replacement. UMBD builds a Bernoulli diffusion model that learns to generate the residual term $U \odot M_{GT}$ while leaving $(1-U)\odot M_c$ untouched, and at inference substitutes a learned uncertainty estimate $\hat U = h(x,M_c)$ from HUQNet for the unknown ground-truth $U$. The authors argue that global noise injection, as in previous diffusion refiners, confuses correctly segmented areas in camouflage scenes, whereas the uncertainty-masked forward process $q(y_t\mid y_0, M_c, U) = \mathrm{B}(U \odot (\bar\alpha_t M_{GT} + (1-\bar\alpha_t) M_c))$ injects noise preferentially where the coarse mask is wrong. The reported consequence is consistent improvement over the coarse masks of SINet, SINetV2, FEDER, Camoformer, and FSNet across four benchmarks, plus gains on polyp and transparent-object segmentation.

Load-bearing premise

The entire refinement gain rests on HUQNet's predicted uncertainty map $\hat U$ being a good surrogate for the true residual $|M_c-M_{GT}|$; if the uncertainty estimate is off, the diffusion process either modifies correct pixels or fails to fix the wrong ones.

Editorial extensions

If this is right

  • Any encoder-decoder COD model can be upgraded by post-processing rather than retraining: with UMBD, SINet, SINetV2, FEDER, Camoformer, and FSNet all improve on all four benchmarks, with average gains of 5.5% in MAE and 3.2% in weighted F-measure.
  • Because the diffusion process models only the uncertain residual, the number of sampling steps can be cut to $T=3$ with little loss of refinement quality, making diffusion-based refinement practical on a single GPU at roughly 0.11 seconds per image at $384\times384$.
  • Global-noise diffusion refiners such as HiDiff and SegRefiner can degrade coarse COD masks by disturbing already-correct regions; UMBD's masking avoids that failure mode and is the only compared refiner that improves the coarse mask on most metrics.
  • The same uncertainty-masked residual formulation transfers to other concealed-object segmentation tasks, reporting substantial gains on polyp segmentation (PraNet, UACANet) and transparent object detection (EBLNet).

Reading between the lines

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

  • The paper reports gains relative to each coarse mask but not the absolute head-to-head with applying the same compute budget to fine-tune the base detector; a natural extension would compare UMBD against equal-cost retraining of the segmenter.
  • A testable prediction follows from the residual identity: if HUQNet's $\hat U$ is replaced by the oracle $U_{GT}$ during inference, the refiner's ceiling should be strictly higher, so measuring that gap would isolate how much of the reported gain comes from uncertainty estimation versus the diffusion mechanism itself.
  • Because the formulation only requires a binary coarse mask and an image, the same uncertainty-masked refiner could plausibly be applied to salient object detection, shadow removal, or scribble-supervised segmentation, where near-binary predictions dominate.
  • The predicted uncertainty map $\hat U$ itself is a by-product that could be used as a failure-localization signal in deployed COD systems, flagging regions where the base detector is unreliable even when refinement is not applied.
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 / 5 minor

Summary. The paper proposes UMBD, an Uncertainty-Masked Bernoulli Diffusion framework for post-processing refinement of coarse masks produced by existing camouflaged object detection (COD) models. The key idea is to estimate a pixel-wise uncertainty map with a Hybrid Uncertainty Quantification Network (HUQNet) and use that map to mask a Bernoulli diffusion process so that the refiner focuses on poorly segmented residual regions while leaving confident regions unchanged. The authors integrate UMBD with six existing COD methods and report average gains of 5.5% in MAE and 3.2% in weighted F-measure across four COD benchmarks, along with additional experiments on polyp segmentation and transparent object detection. The paper includes extensive tables, qualitative comparisons, ablations, and a promised code release.

Significance. If the claims hold, UMBD would be a useful plug-and-play refinement module for COD, with the practical advantage of requiring only a few sampling steps. The paper is also among the first to bring Bernoulli diffusion to COD refinement and to combine it with learned uncertainty masking, which is a plausible and interesting design direction. The breadth of experiments—six base segmenters, four COD benchmarks, and two additional concealed-object tasks—is a strength, and the three-stage training strategy and DDIM acceleration are sensible engineering contributions. However, the significance is conditional on fixing the mathematical issues in the residual formulation: as written, Eq. (1) does not yield the ground-truth mask even with perfect uncertainty, and the forward process in Eqs. (3)-(6) is not a standard Bernoulli diffusion on binary states because y0 is continuous. These problems are load-bearing for the paper's central mechanism, not merely presentational.

major comments (4)
  1. [Sec. 3.1, Eq. (1)] The claimed ideal-refinement property of Eq. (1) is false for continuous Mc. Substituting U=UGT=|Mc-MGT| into Mr=U⊙MGT+(1-U)⊙Mc gives Mr=Mc+|Mc-MGT|(MGT-Mc), which equals MGT only when Mc=MGT or when |Mc-MGT|=1. For example, MGT=1, Mc=0.9 yields Mr=0.91, and MGT=0, Mc=0.2 yields Mr=0.16. This bias propagates to the training target y0=UGT⊙MGT and to Eq. (11): even a perfect denoiser with \hat y0=UGT⊙MGT leaves the term (1-UGT)⊙Mc, so false positives cannot be removed unless U saturates to 1, which training against the soft target UGT does not incentivize. The authors should redefine UGT as a binary error indicator, or reformulate the residual target so that exact recovery actually holds, and then re-derive the training objective accordingly.
  2. [Sec. 3.2, Eqs. (3)-(6)] The forward process is not a well-defined Bernoulli diffusion because y0=U⊙MGT is continuous whenever U is soft, whereas the Bernoulli diffusion framework requires latent states yt in {0,1}. Equation (6) writes yt=y0⊕ϵ with ⊕ denoting XOR, but XOR is not defined for continuous-valued y0 (e.g., U=0.2 and MGT=1 gives y0=0.2). Similarly, Eq. (5) places a Bernoulli distribution on a quantity that is itself continuous through U. The posterior in Eq. (7) and the reparameterization used in training depend on the discrete-state derivation of [43], so this is not a minor notation issue. The authors need to either binarize U/y0, adopt a continuous latent formulation that is compatible with the Bernoulli kernel, or provide a rigorous justification for a mixed discrete-continuous state space.
  3. [Sec. 3.3, Fig. 3] The accuracy of HUQNet's uncertainty estimate \hat U is only demonstrated qualitatively in Fig. 3. Since \hat U is the test-time substitute for UGT and directly controls which regions the diffusion process modifies, the paper should report quantitative agreement between \hat U and UGT (for example, MAE or correlation on a held-out validation split) and should include an ablation in which UGT is supplied at inference instead of \hat U. Without such an evaluation, the attribution of the reported gains to uncertainty-masked residual modeling is not established.
  4. [Sec. 4.2, Tables 1-2] The paper states that all results are averages over five random seeds but reports no standard deviations or significance tests. Several reported gains are very small; for instance, in Table 1, most Sα improvements are below 1%, and some Fβ gains are around 1%. Without error bars or statistical tests, the claim of consistent improvements across methods and datasets is not fully supported. The authors should report variances over the five runs or perform paired significance tests, especially for the small-magnitude gains.
minor comments (5)
  1. [Sec. 2] There is a typo in the Related Work section: 'in the filed' should be 'in the field'.
  2. [Algorithms 1-2 and Eq. (6)] Algorithm 1 obtains yt via XOR in Eq. (6), but the denoising target is written as \hat y0=|yt-\hatϵ|. The relationship between XOR corruption and absolute-difference inversion should be clarified, because the two operations do not coincide for continuous values.
  3. [Table 2] The training or fine-tuning status of HiDiff, SegRefiner, and SAMRefiner on COD data is not stated. Please specify whether released weights were used as-is or whether these baselines were fine-tuned on the same COD training split, so that the comparison is controlled.
  4. [References] References [15] and [40] are duplicate entries for the same Ho et al. paper on denoising diffusion probabilistic models; please consolidate them.
  5. [Fig. 3] The caption refers to columns (e)-(f), but the figure appears to show (e) \hat U and (f) UGT; please make the panel labels explicit in both the caption and the figure.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: UMBD's improvement claim is an empirical, externally-benchmarked result; the residual-formulation inconsistency in Eq. (1) is a correctness issue, not a circular reduction.

full rationale

The paper's central claim is empirical: UMBD refines coarse masks from existing COD segmenters and is evaluated on held-out CHAMELEON/CAMO/COD10K/NC4K benchmarks (Tables 1-2), with MAE and weighted F-measure gains reported against ground truth. The diffusion refiner g(.) is trained with LDiff (KL, IoU, and BCE losses) against MGT, and HUQNet is trained with LH against UGT=|Mc-MGT|; at inference, U_hat=h(x,Mc) feeds the sampler, but the final metrics are compared with external ground truth, so there is no fitted-parameter-renamed-as-prediction loop. Self-citations (e.g., Refs. [10,11,14,16]) are background surveys and prior methods, not load-bearing for the derivation, and no uniqueness or ansatz is imported from the authors' own prior work. The one notable derivation issue is in Sec. 3.1: with UGT=|Mc-MGT|, Eq. (1) gives Mr=|Mc-MGT|*MGT+(1-|Mc-MGT|)*Mc, which is not equal to MGT when Mc is continuous (e.g., MGT=1, Mc=0.9 yields Mr=0.91), so the 'ideal refined mask' claim is numerically false. This is an internal-consistency and correctness flaw, not a case where an output reduces by construction to an input; it does not make the benchmark results circular. Hence no circularity step is scored.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central claim is empirical, not derived from first principles. The method relies on standard diffusion theory and on learnable components (HUQNet, denoiser) with many training hyperparameters. The main modeling choices listed are hand-set, but none constitutes a fitted constant of a physical law.

free parameters (4)
  • eta (η) in BNN loss = 0.1
    Hand-set weight for KL divergence in Eq. (16); authors state it is empirically set to 0.1.
  • Number of inference sampling steps T = 10 default, 3 minimal
    DDIM sub-sequence length; ablation shows T=3 retains performance while T=1 collapses. Chosen by hand to trade quality and speed.
  • Cross-attention window size = 16x16
    Window partitioning size in HUQNet's efficient cross-attention (Sec. 3.3).
  • Head dimension d_h = 4
    Multi-head cross-attention head dimension in HUQNet's fusion module.
assumptions (3)
  • domain assumption The pixel-wise L1 distance UGT=|Mc-MGT| is a valid target for uncertainty-guided refinement and can be predicted from the image and coarse mask.
    Used to define y0 and fMc in the forward process (Eq. 3-5); the method's success relies on HUQNet approximating this quantity at inference.
  • domain assumption Bernoulli diffusion is appropriate for binary mask refinement, and the residual regions are well modeled by a Bernoulli process.
    Core of the method; supported by ablation (Table 4) comparing Gaussian vs. Bernoulli kernels.
  • domain assumption Features from frozen pretrained COD encoders provide useful prior knowledge for the denoiser.
    Feature adaptation in Eq. (12) relies on this transfer.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Uncertainty-Masked Bernoulli Diffusion for Camouflaged Object Detection Refinement." pith.science (2026). https://pith.science/paper/4CHY7GZU

@misc{pith2026250610712,
  author       = {Pith},
  title        = {Pith review of: Uncertainty-Masked Bernoulli Diffusion for Camouflaged Object Detection Refinement},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4CHY7GZU}},
  note         = {Machine review of arXiv:2506.10712}
}
read the original abstract

Camouflaged Object Detection (COD) presents inherent challenges due to the subtle visual differences between targets and their backgrounds. While existing methods have made notable progress, there remains significant potential for post-processing refinement that has yet to be fully explored. To address this limitation, we propose the Uncertainty-Masked Bernoulli Diffusion (UMBD) model, the first generative refinement framework specifically designed for COD. UMBD introduces an uncertainty-guided masking mechanism that selectively applies Bernoulli diffusion to residual regions with poor segmentation quality, enabling targeted refinement while preserving correctly segmented areas. To support this process, we design the Hybrid Uncertainty Quantification Network (HUQNet), which employs a multi-branch architecture and fuses uncertainty from multiple sources to improve estimation accuracy. This enables adaptive guidance during the generative sampling process. The proposed UMBD framework can be seamlessly integrated with a wide range of existing Encoder-Decoder-based COD models, combining their discriminative capabilities with the generative advantages of diffusion-based refinement. Extensive experiments across multiple COD benchmarks demonstrate consistent performance improvements, achieving average gains of 5.5% in MAE and 3.2% in weighted F-measure with only modest computational overhead. Code will be released.

Figures

Figures reproduced from arXiv: 2506.10712 by the authors.

Figure 1
Figure 1. Visual illustration of coarse masks from various COD methods and their corresponding [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. UMBD Framework for COD Refinement. The Bernoulli diffusion-based refiner leverages prior knowledge from pretrained models to perform uncertainty-masked interactive refinement, focusing on residual regions. Red boxes on the samples highlight regions where missing details in MC are progressively attended to and ultimately rectified in Mr during the diffusion process. Panel (b) presents the proposed HUQNet with detaile… view at source ↗
Figure 3
Figure 3. Qualitative refinement results by our UMBD with uncertainty estimates [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Qualitative comparisons with other refiners on COD benchmarks. Blue/yellow backgrounds [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Qualitative results by our UMBD on polyp segmentation. Please zoom in for a better view. [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]
Figure 6
Figure 6. Figure 6: Qualitative refinement results on transparent object detection. Zoom in for a better view. [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: Qualitative comparisons with other refiners on COD benchmarks. Blue/yellow backgrounds [PITH_FULL_IMAGE:figures/full_fig_p015_7.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Refining Context-Entangled Content Segmentation via Curriculum Selection and Anti-Curriculum Promotion

    cs.CV 2026-02 conditional novelty 6.0 of 10

    A curriculum-then-anti-curriculum training schedule, ending with spectral low-pass fine-tuning, improves context-entangled segmentation across several datasets and backbones.

  2. Discrete Diffusion Models: A Unified Framework from Tokenization to Generation

    cs.LG 2026-07 unverdicted novelty 4.0 of 10

    Discrete diffusion models are re-framed as instances of a tokenization-centric, four-component design space (corruption, denoiser, objective, sampler) in a broad survey with no new experimental or theoretical results.

Reference graph

Works this paper leans on

71 extracted references · 48 canonical work pages · cited by 2 Pith papers

  1. [43]

    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,” inNeural Information Processing Systems, 2021

    work page 2021

  2. [1]

    Concealed object detection,

    D.-P. Fan, G.-P. Ji, M.-M. Cheng, and L. Shao, “Concealed object detection,”IEEE transactions on pattern analysis and machine intelligence, vol. 44, no. 10, pp. 6024–6042, 2021

    work page 2021

  3. [2]

    Deep Learning in Concealed Dense Prediction

    P. Zhao, D.-P. Fan, S. Cheng, S. Khan, F. S. Khan, D. Clifton, P. Xu, and J. Yang, “Deep learning in concealed dense prediction,”arXiv preprint arXiv:2504.10979, 2025

    work page Pith review arXiv 2025

  4. [3]

    Depth-aware concealed crop detection in dense agricultural scenes,

    L. Wang, J. Yang, Y . Zhang, F. Wang, and F. Zheng, “Depth-aware concealed crop detection in dense agricultural scenes,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 17 201–17 211

    work page 2024

  5. [5]

    Computer-vision-based fabric defect detection: A survey,

    A. Kumar, “Computer-vision-based fabric defect detection: A survey,”IEEE transactions on industrial electronics, vol. 55, no. 1, pp. 348–363, 2008

    work page 2008

  6. [6]

    Image threshold segmentation based on glle histogram,

    C. He, X. Wang, L. Deng, and G. Xu, “Image threshold segmentation based on glle histogram,” in2019 International Conference on Internet of Things (iThings) and IEEE Green Computing and Communica- tions (GreenCom) and IEEE Cyber, Physical and Social Computing (CPSCom) and IEEE Smart Data (SmartData). IEEE, 2019, pp. 410–415

    work page 2019

  7. [7]

    Camouflaged object detection,

    D.-P. Fan, G.-P. Ji, G. Sun, M.-M. Cheng, J. Shen, and L. Shao, “Camouflaged object detection,” inIEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2020

    work page 2020

  8. [8]

    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 Transactions on Pattern Analysis and Machine Intelligence, 2024

    2024

Show all 71 references
  1. [9]

    Camouflaged object detection with feature decomposition and edge reconstruction,

    C. He, K. Li, Y . Zhang, L. Tang, Y . Zhang, Z. Guo, and X. Li, “Camouflaged object detection with feature decomposition and edge reconstruction,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2023, pp. 22 046–22 055

    2023

  2. [10]

    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,”arXiv preprint arXiv:2501.18783, 2025

    2025 arXiv

  3. [11]

    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

    2025 arXiv

  4. [12]

    Tackling the over-smoothing problem of cnn-based hyperspectral image classification,

    S. He, H. Xue, J. Cheng, L. Wang, Y . Wang, and Y . Zhang, “Tackling the over-smoothing problem of cnn-based hyperspectral image classification,”Journal of Applied Remote Sensing, vol. 16, no. 4, pp. 048 506–048 506, 2022

    2022

  5. [13]

    Mitigating undisciplined over-smoothing in transformer for weakly supervised semantic segmentation,

    J. He, L. Cheng, C. Fang, D. Zhang, Z. Wang, and W. Chen, “Mitigating undisciplined over-smoothing in transformer for weakly supervised semantic segmentation,”arXiv preprint arXiv:2305.03112, 2023

    2023 arXiv

  6. [14]

    A survey of camouflaged object detection and beyond,

    F. Xiao, S. Hu, Y . Shen, C. Fang, J. Huang, C. He, L. Tang, Z. Yang, and X. Li, “A survey of camouflaged object detection and beyond,”arXiv preprint arXiv:2408.14562, 2024

    2024 arXiv

  7. [15]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,”Advances in neural information processing systems, vol. 33, pp. 6840–6851, 2020

    2020

  8. [16]

    Diffusion models in low-level vision: A survey,

    C. He, Y . Shen, C. Fang, F. Xiao, L. Tang, Y . Zhang, W. Zuo, Z. Guo, and X. Li, “Diffusion models in low-level vision: A survey,”arXiv preprint arXiv:2406.11138, 2024

    2024 arXiv

  9. [17]

    Diffusion models in vision: A survey,

    F.-A. Croitoru, V . Hondru, R. T. Ionescu, and M. Shah, “Diffusion models in vision: A survey,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 9, pp. 10 850–10 869, 2023

    2023

  10. [18]

    Camodiffusion: Camouflaged object detection via conditional diffusion models,

    Z. Chen, K. Sun, and X. Lin, “Camodiffusion: Camouflaged object detection via conditional diffusion models,” inProceedings of the AAAI Conference on Artificial Intelligence, vol. 38, no. 2, 2024, pp. 1272–1280

    2024

  11. [19]

    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

    2025

  12. [20]

    Uncertainty-guided diffusion model for camouflaged object detection,

    J. Yang, B. Zhong, Q. Liang, Z. Mo, S. Zhang, and S. Song, “Uncertainty-guided diffusion model for camouflaged object detection,”IEEE Transactions on Multimedia, 2025. 10

    2025

  13. [21]

    Diffusion model for camouflaged object detection,

    Z. Chen, R. Gao, T.-Z. Xiang, and F. Lin, “Diffusion model for camouflaged object detection,” inECAI

  14. [22]

    A brief survey on semantic segmentation with deep learning,

    S. Hao, Y . Zhou, and Y . Guo, “A brief survey on semantic segmentation with deep learning,”Neurocomput- ing, vol. 406, pp. 302–321, 2020

    2020

  15. [23]

    Deep unsupervised learning using nonequilibrium thermodynamics,

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

    2015

  16. [24]

    Structured denoising diffusion models in discrete state-spaces,

    J. Austin, D. D. Johnson, J. Ho, D. Tarlow, and R. van den Berg, “Structured denoising diffusion models in discrete state-spaces,” inNeurIPS, 2021, pp. 17 981–17 993

    2021

  17. [25]

    Hidiff: hybrid diffusion framework for medical image segmentation,

    T. Chen, C. Wang, Z. Chen, Y . Lei, and H. Shan, “Hidiff: hybrid diffusion framework for medical image segmentation,”IEEE Transactions on Medical Imaging, 2024

    2024

  18. [26]

    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,”arXiv preprint arXiv:2312.12425, 2023

    2023 arXiv

  19. [27]

    Pointrend: Image segmentation as rendering,

    A. Kirillov, Y . Wu, K. He, and R. Girshick, “Pointrend: Image segmentation as rendering,” inCVPR, 2020, pp. 9799–9808

    2020

  20. [28]

    Mask transfiner for high-quality instance segmentation,

    L. Ke, M. Danelljan, X. Li, Y .-W. Tai, C.-K. Tang, and F. Yu, “Mask transfiner for high-quality instance segmentation,” inCVPR, 2022, pp. 4412–4421

    2022

  21. [29]

    Look closer to segment better: Boundary patch refinement for instance segmentation,

    C. Tang, H. Chen, X. Li, J. Li, Z. Zhang, and X. Hu, “Look closer to segment better: Boundary patch refinement for instance segmentation,” inCVPR, 2021, pp. 13 926–13 935

    2021

  22. [30]

    Segfix: Model-agnostic boundary refinement for segmentation,

    Y . Yuan, J. Xie, X. Chen, and J. Wang, “Segfix: Model-agnostic boundary refinement for segmentation,” in ECCV. Springer, 2020, pp. 489–506

    2020

  23. [31]

    Cascadepsp: Toward class-agnostic and very high- resolution segmentation via global and local refinement,

    H. K. Cheng, J. Chung, Y .-W. Tai, and C.-K. Tang, “Cascadepsp: Toward class-agnostic and very high- resolution segmentation via global and local refinement,” inCVPR, 2020, pp. 8890–8899

    2020

  24. [32]

    Polytransform: Deep polygon transformer for instance segmentation,

    J. Liang, N. Homayounfar, W.-C. Ma, Y . Xiong, R. Hu, and R. Urtasun, “Polytransform: Deep polygon transformer for instance segmentation,” inCVPR, 2020, pp. 9131–9140

    2020

  25. [33]

    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,”arXiv preprint arXiv:2502.06756, 2025

    2025 arXiv

  26. [34]

    Segment anything,

    A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y . Loet al., “Segment anything,” inProceedings of the IEEE/CVF international conference on computer vision, 2023, pp. 4015–4026

    2023

  27. [35]

    Predictive uncertainty estimation for camouflaged object detection,

    Y . Zhang, J. Zhang, W. Hamidouche, and O. Deforges, “Predictive uncertainty estimation for camouflaged object detection,”IEEE Transactions on Image Processing, vol. 32, pp. 3580–3591, 2023

    2023

  28. [36]

    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,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 4146–4155

    2021

  29. [37]

    Uncertainty-aware joint salient object and camouflaged object detection,

    A. Li, J. Zhang, Y . Lv, B. Liu, T. Zhang, and Y . Dai, “Uncertainty-aware joint salient object and camouflaged object detection,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2021, pp. 10 071–10 081

    2021

  30. [38]

    Preynet: Preying on camouflaged objects,

    M. Zhang, S. Xu, Y . Piao, D. Shi, S. Lin, and H. Lu, “Preynet: Preying on camouflaged objects,” in Proceedings of the 30th ACM international conference on multimedia, 2022, pp. 5323–5332

    2022

  31. [39]

    Modeling aleatoric uncertainty for camouflaged object detection,

    J. Liu, J. Zhang, and N. Barnes, “Modeling aleatoric uncertainty for camouflaged object detection,” in Proceedings of the IEEE/CVF winter conference on applications of computer vision, 2022, pp. 1445–1454

    2022

  32. [40]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,”NeurIPS, pp. 6840–6851, 2020

    2020

  33. [41]

    Generative modeling by estimating gradients of the data distribution,

    Y . Song and S. Ermon, “Generative modeling by estimating gradients of the data distribution,”NeurIPS, 2019

    2019

  34. [42]

    U-net: Convolutional networks for biomedical image segmenta- tion,

    O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image segmenta- tion,” inMedical image computing and computer-assisted intervention–MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part III 18. ...

    2015

  35. [44]

    Denoising diffusion implicit models,

    J. Song, C. Meng, and S. Ermon, “Denoising diffusion implicit models,”arXiv preprint arXiv:2010.02502, 2020

    2010 arXiv

  36. [45]

    Decoupled weight decay regularization,

    I. Loshchilov and F. Hutter, “Decoupled weight decay regularization,” inICLR, 2017

    2017

  37. [46]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 770–778

    2016

  38. [47]

    Imagenet: A large-scale hierarchical image database,

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, “Imagenet: A large-scale hierarchical image database,” in2009 IEEE conference on computer vision and pattern recognition. Ieee, 2009, pp. 248–255

    2009

  39. [48]

    Improved denoising diffusion probabilistic models,

    A. Nichol and P. Dhariwal, “Improved denoising diffusion probabilistic models,”ArXiv, vol. abs/2102.09672, 2021

    2021 arXiv

  40. [49]

    Animal camouflage analysis: Chameleon database,

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

    2018

  41. [50]

    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

    2019

  42. [51]

    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,” inCVPR, 2021, pp. 11 591–11 601

    2021

  43. [52]

    How to evaluate foreground maps?

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

    2014

  44. [53]

    Cognitive vision inspired object segmentation metric and loss function,

    D.-P. Fan, G.-P. Ji, X. Qin, and M.-M. Cheng, “Cognitive vision inspired object segmentation metric and loss function,”Scientia Sinica Informationis, vol. 6, no. 6, 2021

    2021

  45. [54]

    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,” inICCV, 2017, pp. 4548–4557

    2017

  46. [55]

    Camouflaged object segmentation based on matching–recognition– refinement network,

    X. Yan, M. Sun, Y . Han, and Z. Wang, “Camouflaged object segmentation based on matching–recognition– refinement network,”IEEE Transactions on Neural Networks and Learning Systems, 2023

    2023

  47. [56]

    Fsnet: Focus scanning network for camouflaged object detection,

    Z. Song, X. Kang, X. Wei, H. Liu, R. Dian, and S. Li, “Fsnet: Focus scanning network for camouflaged object detection,”IEEE Transactions on Image Processing, vol. 32, pp. 2267–2278, 2023

    2023

  48. [57]

    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 transactions on pattern analysis and machine intelligence, vol. 43, no. 2, pp. 652–662, 2019

    2019

  49. [58]

    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,”Computational visual media, vol. 8, no. 3, pp. 415–424, 2022

    2022

  50. [59]

    Sam-adapter: Adapting segment anything in underperformed scenes,

    T. Chen, L. Zhu, C. Deng, R. Cao, Y . Wang, S. Zhang, Z. Li, L. Sun, Y . Zang, and P. Mao, “Sam-adapter: Adapting segment anything in underperformed scenes,” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 3367–3375

    2023

  51. [60]

    Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps,

    C. Lu, Y . Zhou, F. Bao, J. Chen, C. Li, and J. Zhu, “Dpm-solver: A fast ode solver for diffusion probabilistic model sampling in around 10 steps,”Advances in Neural Information Processing Systems, vol. 35, pp. 5775–5787, 2022

    2022

  52. [61]

    A survey on deep learning for polyp segmentation: Techniques, challenges and future trends,

    J. Mei, T. Zhou, K. Huang, Y . Zhang, Y . Zhou, Y . Wu, and H. Fu, “A survey on deep learning for polyp segmentation: Techniques, challenges and future trends,”Visual Intelligence, vol. 3, no. 1, p. 1, 2025

    2025

  53. [62]

    Confusing object detection: A survey

    K. Tong, G. Zou, X. Tan, J. Gong, Z. Qi, Z. Zhang, Y . Xie, and L. Ma, “Confusing object detection: A survey.”Computers, Materials & Continua, vol. 80, no. 3, 2024

    2024

  54. [63]

    Wm-dova maps for accurate polyp highlighting in colonoscopy: Validation vs. saliency maps from physicians,

    J. Bernal, F. J. Sánchez, G. Fernández-Esparrach, D. Gil, C. Rodríguez, and F. Vilariño, “Wm-dova maps for accurate polyp highlighting in colonoscopy: Validation vs. saliency maps from physicians,” Computerized medical imaging and graphics, vol. 43, pp. 99–111, 2015

    2015

  55. [64]

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

    J. Silva, A. Histace, O. Romain, X. Dray, and B. Granado, “Toward embedded detection of polyps in wce images for early diagnosis of colorectal cancer,”International journal of computer assisted radiology and surgery, vol. 9, pp. 283–293, 2014. 12

    2014

  56. [65]

    Pranet: Parallel reverse attention network for polyp segmentation,

    D.-P. Fan, G.-P. Ji, T. Zhou, G. Chen, H. Fu, J. Shen, and L. Shao, “Pranet: Parallel reverse attention network for polyp segmentation,” inInternational conference on medical image computing and computer-assisted intervention. Springer, 2020, pp. 263–273

    2020

  57. [66]

    Uacanet: Uncertainty augmented context attention for polyp segmentation,

    T. Kim, H. Lee, and D. Kim, “Uacanet: Uncertainty augmented context attention for polyp segmentation,” Proceedings of the 29th ACM International Conference on Multimedia, 2021

    2021

  58. [67]

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

    H. Mei, X. Yang, Y . Wang, Y . Liu, S. He, Q. Zhang, X. Wei, and R. W. Lau, “Don’t hit me! glass detection in real-world scenes,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 3687–3696

    2020

  59. [68]

    Where is my mirror?

    X. Yang, H. Mei, K. Xu, X. Wei, B. Yin, and R. W. Lau, “Where is my mirror?” inProceedings of the IEEE/CVF International Conference on Computer Vision, 2019, pp. 8809–8818

    2019

  60. [69]

    Enhanced boundary learning for glass-like object segmentation,

    H. He, X. Li, G. Cheng, J. Shi, Y . Tong, G. Meng, V . Prinet, and L. Weng, “Enhanced boundary learning for glass-like object segmentation,” inProceedings of the IEEE/CVF international conference on computer vision, 2021, pp. 15 859–15 868

    2021

  61. [70]

    Feature shrinkage pyramid for camouflaged object detection with transformers,

    Z. Huang, H. Dai, T.-Z. Xiang, S. Wang, H.-X. Chen, J. Qin, and H. Xiong, “Feature shrinkage pyramid for camouflaged object detection with transformers,” inIEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023

    2023

  62. [71]

    Focusdiffuser: Perceiving local disparities for camouflaged object detection,

    J. Zhao, X. Li, F. Yang, Q. Zhai, A. Luo, Z. Jiao, and H. Cheng, “Focusdiffuser: Perceiving local disparities for camouflaged object detection,” inEuropean Conference on Computer Vision. Springer, 2024, pp. 181–198. A Appendix A.1 Efficiency Analysis Table 6: Impact of reducin...

    2024

  63. [2023]

    IOS Press, 2023, pp. 445–452

    2023

Pith tools

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