Pith. sign in

REVIEW 4 major objections 6 minor 69 references

UCOD-DPL: Unsupervised Camouflaged Object Detection via Dynamic Pseudo-label Learning

T0 review · 4 major / 6 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 The paper claims that an entirely unsupervised teacher-student pipeline, with a dynamically mixed pseudo-label and a zoom-in second pass, outperforms all prior unsupervised camouflaged-object detectors and several supervised ones.

desk verdict Solid unsupervised COD paper with a real gain over prior UCOD methods; the main risk is pseudo-label bootstrapping locking in seed errors, which needs a corruption-injection test, but it deserves a serious referee. read the letter →

arxiv 2506.07087 v1 pith:UYJOINZR submitted 2025-06-08 cs.CV

Weiqi Yan , Lvhai Chen , Huaijia Kou , Shengchuan Zhang , Yan Zhang , Liujuan Cao This is my paper
classification cs.CV
keywords unsupervisedcamouflagedobjectdetectionpseudo-labellearningteacher-studentframeworkadaptivemixingdual-branchadversarialdecoderlook-twicerefinementDINOv2semanticsegmentation
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

This paper sets out to show that camouflaged object detection can be done without any pixel-level human labels and still reach the level of several fully supervised methods. The authors identify two reasons earlier unsupervised attempts lag: fixed-strategy pseudo-labels are noisy, and a bare $1\times1$ convolutional decoder cannot separate camouflaged foreground from background, especially for small objects. Their UCOD-DPL answers with a teacher-student framework whose Adaptive Pseudo-label Module mixes the fixed-strategy label with the teacher's prediction according to a time-varying weight, a Dual-Branch Adversarial decoder that learns foreground and background features separately with an orthogonality penalty, and a Look-Twice mechanism that crops and re-segments small predicted objects. Across the CHAMELEON, CAMO, COD10K-Test, and NC4K benchmarks, the paper reports that the method outperforms all prior unsupervised baselines and several semi- and fully-supervised competitors.

What carries the argument

The load-bearing mechanism is the Adaptive Pseudo-label Module (APM). A discriminator $D$ scores the fixed-strategy pseudo-label $\hat P^{fs}_i$ and the student's foreground prediction $\hat Y^{FG}_i$, and a cosine schedule converts the difference into a mixing weight $W_i^t = S(\hat y^{p1}_i,\hat y^{p2}_i)$, so the training label $P_i = W_i^t \hat P_i^t + (1-W_i^t)\hat P_i^{fs}$ starts by trusting the fixed strategy and later trusts the teacher. Two further mechanisms carry the argument: the Dual-Branch Adversarial (DBA) decoder, which splits features into foreground and background branches with an orthogonality loss $L_\perp$ so the two tasks cannot share the same features, and the Look-Twice mechanism, which labels connected components of the coarse mask, crops those below a size threshold $\tau=0.15$, re-infers them at full resolution, and pastes the refined masks back.

What would settle it

Compare the background-seed pseudo-labels against human ground truth on the COD10K-Test set, separately for small-object images; if the seed's overlap with the ground truth is no better than chance on those images, then the reported gains cannot originate from the fixed strategy, and the APM correction plus the decoder would be doing all the work.

Watch

Extended reading notes

Core claim

The central claim is that the training signal for unsupervised camouflaged object detection should be a dynamic, self-correcting label rather than a fixed noisy mask. The paper's specific discovery is that a discriminator can tell when the student's prediction has collapsed onto the fixed strategy, and a cosine-scored mixing weight can then shift supervisory trust from the fixed background-seed pseudo-label to the teacher's pseudo-label as training proceeds, letting the model learn from the seed without memorizing its errors. On top of this, the Dual-Branch Adversarial decoder separates foreground and background into parallel attention branches with an orthogonality loss, and the Look-Twice mechanism re-infers small connected components at higher resolution. The paper claims this combination surpasses all existing unsupervised and semi-supervised methods on the four standard test sets and exceeds several fully supervised methods when DINOv2 is the backbone.

Load-bearing premise

The load-bearing premise is that the background-seed pseudo-label computed from frozen DINOv2 features is good enough to bootstrap learning; because the teacher is only a slowly updated copy of the student, systematic errors in that seed can be inherited by the teacher and become the very target the student is trained to fit.

Editorial extensions

If this is right

  • If the central claim holds, pixel-level annotation is not a prerequisite for practical camouflage detection: a frozen self-supervised backbone plus unlabeled images is enough to reach the level of several trained-with-labels detectors.
  • The APM weighting rule—high trust in the fixed strategy early, high trust in the teacher later—is a general cure for confirmation bias in self-training, applicable to any noisy pseudo-label pipeline.
  • The DBA decoder gives a concrete decoder design for other binary segmentation tasks where foreground and background are easily confused: separate branches plus an orthogonality penalty.
  • Because the Look-Twice pass drives the small-object gains, the method's advantage over prior unsupervised baselines should be largest on datasets dominated by small camouflaged targets.

Reading between the lines

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

  • A natural test the paper does not run is replacing the background-seed fixed strategy with other unsupervised proposals (spectral clustering or MaskCut) and checking whether APM still converges; the ablation only varies noise and solid-color seeds, not the seed algorithm.
  • The reported edge over fully supervised methods may owe as much to the DINOv2 backbone as to the UCOD pipeline; matching supervised decoders on the same frozen backbone would isolate the contribution.
  • The paper's own acknowledged weakness—Look-Twice can amplify a fragment of an occluded object as a small object—suggests the gains on small objects will not transfer to heavily occluded scenes until the connected-component decomposition becomes occlusion-aware.
  • The same dynamic-mixing idea could apply to medical or remote-sensing segmentation, where pseudo-labels are noisy, foreground is tiny, and labeled data are scarce; the paper does not test those domains.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper proposes UCOD-DPL, an unsupervised camouflaged object detection method built around a teacher-student framework with an Adaptive Pseudo-label Mixing (APM) module, a Dual-Branch Adversarial (DBA) decoder, and a Look-Twice refinement mechanism. The APM mixes a fixed-strategy pseudo-label (FOUND background seed) with the teacher prediction using a discriminator-derived score plus an epoch ramp; the DBA decoder splits features into foreground and background branches with an orthogonal loss; Look-Twice crops and re-infers small connected components. The method is trained on unlabeled CAMO-Train and COD10K-Train images and evaluated on CHAMELEON, CAMO-Test, COD10K-Test, and NC4K, reporting SOTA results among USS/UCOD methods and competitive or better results than several fully supervised methods. The paper includes module and strategy ablations, a foreground-size analysis, and a code link.

Significance. If the reported results hold, the paper makes a meaningful empirical contribution: it shows that an unsupervised COD method can approach the performance of fully supervised COD models on four established benchmarks, and it releases code. The ablations are useful and each proposed module gives a clear improvement in Table 2. The paper is also candid about a limitation of Look-Twice for fragmented objects. However, the central mechanism advertised as preventing error accumulation is not directly validated, and at least one hyperparameter is selected on the test set, so the strength of the SOTA claim currently rests on incomplete evidence.

major comments (4)
  1. [Sec. 4.3, Eq. (2)] The claim that APM adaptively mixes the fixed-strategy and teacher pseudo-labels is undermined by the structure of Eq. (2). Since S = CLIP(t/T + 0.5(1+cos(pi*|yp1-yp2|))), the deterministic ramp t/T dominates: for t/T >= 0.5 the cosine term is at most 0.5, so W_i^t >= t/T; at t = T the weight is identically 1. Thus in the second half of training the discriminator-dependent term cannot reduce the teacher weight below 0.5, and at the final epoch the mixing is purely teacher-based for every image. The paper does not report the distribution of W_i^t over training, so the adaptive behavior claimed in Sec. 3.2 is not demonstrated. Please provide per-epoch statistics of W_i^t and an ablation that removes the t/T term (or uses only the discriminator score) to show that the adaptive component, rather than the schedule, drives the gain in Table 3.
  2. [Sec. 3.4 and Sec. 4.3 (Fig. 5)] The hyperparameter tau (small-object ratio threshold) is selected by an ablation on COD10K-Test, the same 2,026-image test set used for the headline results in Table 1. This constitutes test-set peeking and can inflate the reported metrics for Look-Twice; the paper does not mention a validation split. Please select tau on a held-out portion of the training set (or perform cross-validation) and report the resulting test numbers, or clearly state that the reported numbers involve test-set model selection.
  3. [Sec. 3.2, Eq. (3), and Table 4] The central claim that APM 'prevent[s] the model from overfitting incorrect knowledge' is not directly tested. In Eq. (3), the training target Pi is a convex combination of the fixed FOUND background-seed pseudo-label and the teacher prediction, where the teacher is an EMA of the student; Table 4 shows only that performance degrades when the seed is replaced by random noise, which establishes that the seed matters but does not establish that the APM-teacher loop corrects systematic seed errors rather than propagating them. Please add a corruption-injection study (e.g., artificially remove or mislabel a known fraction of foreground regions in the seed and measure whether the final predictions recover the missing regions, comparing APM against the fixed seed alone and against 1:1 mixing).
  4. [Sec. 4.2, Table 1] The SOTA claim relies on '*FOUND23-DINOv2' and '*UCOS-DA23-DINOv2', which are reimplemented by the authors. No details are given for these reimplementations (architecture, training schedule, post-processing), so a reader cannot verify that they are faithful. Since the paper reports beating these reimplemented baselines, please provide the reimplementation protocol, or use the original published numbers with DINOv1 and clearly state the comparison setting.
minor comments (6)
  1. [Abstract and Sec. 1] The word 'Camoflaged' is misspelled; it should be 'Camouflaged'.
  2. [Sec. 4.1 and Table 1] Sec. 4.1 states CHAMELEON has 76 test images, but Table 1 header reports 87; please correct the inconsistency.
  3. [Sec. 3.3] The teacher pseudo-label is denoted \hat P^t_i in Sec. 3.1, but the last paragraph of Sec. 3.3 refers to it as Pi; please use consistent notation.
  4. [Eq. (11)] The text says 'H, W denote the height and width of the foreground area of the output mask'; this should read 'of the output mask', since the ratio is normalized by the full mask size.
  5. [Sec. 3.4, Eq. (12)] The definitions of s_FG_k and s_BG_k are easy to confuse; consider renaming s_BG_k to 'bounding-box area ratio' and clarifying the notation.
  6. [Fig. 5] The axis labels use 'Foreground Ratio r' while the caption refers to tau; clarify the relationship between r and tau.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity is present: the SOTA claim is evaluated on held-out ground truth, and the teacher-student pseudo-label loop is self-referential but not definitionally circular.

full rationale

The claimed derivation does not reduce to its own inputs. The training target Pi in Eq. (3) is a convex combination of the fixed-strategy pseudo-label and the EMA-teacher prediction, and the mixing weight W_t in Eq. (2) follows a deterministic epoch ramp; this is a self-training loop, but the reported SOTA claim in Sec. 4.2 is measured against held-out ground truth on four external test sets, so the headline numbers are not constructed from the training targets. The discriminator in Eq. (4) distinguishes fixed-strategy masks from student masks rather than judging mask correctness, which weakens the paper's self-correction narrative but is not a circular step. The tau threshold for Look-Twice is tuned on COD10K-Test in Sec. 4.3, which is a test-set selection concern, but no equation forces the reported metric from that choice, so it does not meet the bar for definitional circularity. The paper's self-citations (e.g., CamoTeacher [27] and Active Teacher [33]) appear as comparison baselines or related-work examples, and no load-bearing argument depends on them. Accordingly, no circular step meets the quoted-evidence requirement, and the central benchmark claim remains externally checkable.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The system imports strong priors: DINOv2 features, a fixed background-seed segmenter, EMA teacher reliability, and a small-object heuristic. The only task-specific free parameter explicitly fitted is tau, tuned directly on the test split; no commit hash or supplementary file is shipped in the paper.

free parameters (2)
  • tau (small-object ratio threshold) = 0.15
    Chosen by hyper-parameter ablation on COD10K-Test (Sec. 4.3, Fig. 5), then used for the final results on the same test set; this is a test-set-tuned free parameter.
  • EMA momentum eta = 0.99
    Set by hand for the teacher update (Sec. 4.1); standard practice but not derived, and it controls how strongly teacher pseudo-labels lag the student.
assumptions (5)
  • domain assumption DINOv1/DINOv2 features encode sufficient semantic and localization information for camouflaged objects.
    Used as the frozen encoder and feature source throughout (Sec. 2.3, Sec. 3.1); if these features miss camouflage, the teacher and student cannot recover.
  • domain assumption The FOUND background-seed pseudo-label is a useful, if noisy, foreground prior for camouflaged objects.
    Used as the untrainable fixed strategy and one of the two mixing inputs in Eq. (3); systematic errors in this seed bound what the model can learn.
  • domain assumption EMA-updated teacher pseudo-labels become more reliable than the fixed strategy over training.
    The entire teacher-student design assumes teacher predictions are a better target late in training (Sec. 3.2); no experiment in the paper isolates teacher-quality improvement from student overfitting.
  • domain assumption The discriminator confidence difference is a meaningful reliability signal for mixing.
    APM weights in Eq. (2) are built from the discriminator's distinction between fixed-strategy and student masks; if the discriminator exploits trivial differences, the mixing weights lose meaning.
  • ad hoc to paper Connected components of the coarse foreground mask correspond to individual objects for zooming.
    Look-Twice (Sec. 3.4, Eqs. 10-13) crops and refines every component smaller than tau; the paper itself admits in Sec. 3.4 that fragmented or occluded objects can be wrongly amplified.

how reviews work

0 comments
Cite this review

Pith. "Pith review of UCOD-DPL: Unsupervised Camouflaged Object Detection via Dynamic Pseudo-label Learning." pith.science (2026). https://pith.science/paper/UYJOINZR

@misc{pith2026250607087,
  author       = {Pith},
  title        = {Pith review of: UCOD-DPL: Unsupervised Camouflaged Object Detection via Dynamic Pseudo-label Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UYJOINZR}},
  note         = {Machine review of arXiv:2506.07087}
}
read the original abstract

Unsupervised Camoflaged Object Detection (UCOD) has gained attention since it doesn't need to rely on extensive pixel-level labels. Existing UCOD methods typically generate pseudo-labels using fixed strategies and train 1 x1 convolutional layers as a simple decoder, leading to low performance compared to fully-supervised methods. We emphasize two drawbacks in these approaches: 1). The model is prone to fitting incorrect knowledge due to the pseudo-label containing substantial noise. 2). The simple decoder fails to capture and learn the semantic features of camouflaged objects, especially for small-sized objects, due to the low-resolution pseudo-labels and severe confusion between foreground and background pixels. To this end, we propose a UCOD method with a teacher-student framework via Dynamic Pseudo-label Learning called UCOD-DPL, which contains an Adaptive Pseudo-label Module (APM), a Dual-Branch Adversarial (DBA) decoder, and a Look-Twice mechanism. The APM module adaptively combines pseudo-labels generated by fixed strategies and the teacher model to prevent the model from overfitting incorrect knowledge while preserving the ability for self-correction; the DBA decoder takes adversarial learning of different segmentation objectives, guides the model to overcome the foreground-background confusion of camouflaged objects, and the Look-Twice mechanism mimics the human tendency to zoom in on camouflaged objects and performs secondary refinement on small-sized objects. Extensive experiments show that our method demonstrates outstanding performance, even surpassing some existing fully supervised methods. The code is available now.

Figures

Figures reproduced from arXiv: 2506.07087 by the authors.

Figure 1
Figure 1. Comparison between our method and previous pseudo [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Some examples of low-quality pseudo-labels generated using fixed strategies. Based on features extracted by DINOv2, we construct pseudo-labels using the Background-Seed from FOUND [44] and the MaskCut in UCOD-DA [44], in challenging scenarios (e.g. small-sized objects, huge-sized objects, multiple instances, and uncertain objects) for visualization. strategies contains substantial noise, which makes the model prone … view at source ↗
Figure 3
Figure 3. The main framework of our proposed method. The model contains a teacher-student framework, an Adaptive Psdudo-label Mixing (APM) module, a Dual-Branch Adversarial (DBA) decoder, and a Look-Twice strategy. not yet developed sufficient localization and basic semantic segmentation capabilities for camouflaged objects, making it unable to supervise the student model effectively. At this point, there is a significant dif… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Visual comparison of our method with other existing methods in challenging scenarios. Our method has clearer and more precise segmentation boundaries and correctly recognizes depth-artifacted objects. results show that our model outperformed all existing USS and UCOD m…
Figure 6
Figure 6. Figure 6: Performance comparison for different foreground sizes on COD10K-Test dataset. black and white masks to replace the fixed-strategy pseudo￾label during the training process. The results are shown in Tab. 4. When using random noise, the noise may over￾lap with the foregro…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

69 extracted references · 64 canonical work pages

  1. [1]

    Object discovery with a copy-pasting gan

    Relja Arandjelovi´c and Andrew Zisserman. Object discovery with a copy-pasting gan. arXiv preprint arXiv:1905.11369,

    arXiv 1905

  2. [2]

    Beit: Bert pre-training of image transformers

    Hangbo Bao, Li Dong, Songhao Piao, and Furu Wei. Beit: Bert pre-training of image transformers. arXiv preprint arXiv:2106.08254, 2021. 3

    arXiv 2021

  3. [3]

    Emergence of object segmen- tation in perturbed generative models

    Adam Bielski and Paolo Favaro. Emergence of object segmen- tation in perturbed generative models. Advances in Neural Information Processing Systems, 32, 2019. 2

    work page 2019

  4. [4]

    The adaptive significance of coloration in mam- mals

    TIM Caro. The adaptive significance of coloration in mam- mals. BioScience, 55(2):125–136, 2005. 1

    work page 2005

  5. [5]

    Emerg- ing properties in self-supervised vision transformers

    Mathilde Caron, Hugo Touvron, Ishan Misra, Hervé Jégou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerg- ing properties in self-supervised vision transformers. In The IEEE/CVF International Conference on Computer Vision , pages 9650–9660, 2021. 3, 6

    work page 2021

  6. [6]

    A simple framework for contrastive learning of visual representations

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Ge- offrey Hinton. A simple framework for contrastive learning of visual representations. In International conference on machine learning, pages 1597–1607. PMLR, 2020. 3

    work page 2020

  7. [7]

    Global contrast based salient region detection

    Ming-Ming Cheng, Niloy J Mitra, Xiaolei Huang, Philip HS Torr, and Shi-Min Hu. Global contrast based salient region detection. IEEE transactions on pattern analysis and machine intelligence, 37(3):569–582, 2014. 2

    work page 2014

  8. [8]

    Structure-measure: A new way to evaluate fore- ground maps

    Deng-Ping Fan, Ming-Ming Cheng, Yun Liu, Tao Li, and Ali Borji. Structure-measure: A new way to evaluate fore- ground maps. In The IEEE/CVF International Conference on Computer Vision, pages 4548–4557, 2017. 6

    work page 2017

Show all 69 references
  1. [9]

    Enhanced-alignment measure for binary foreground map evaluation

    Deng-Ping Fan, Cheng Gong, Yang Cao, Bo Ren, Ming- Ming Cheng, and Ali Borji. Enhanced-alignment measure for binary foreground map evaluation. In Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence, pages 698–704. AAAI Press, 2018. 6

    2018

  2. [10]

    Camouflaged object detection

    Deng-Ping Fan, Ge-Peng Ji, Guolei Sun, Ming-Ming Cheng, Jianbing Shen, and Ling Shao. Camouflaged object detection. In The IEEE/CVF Computer Vision and Pattern Recognition Conference, pages 2777–2787, 2020. 3, 6

    2020

  3. [11]

    Pranet: Parallel reverse at- tention network for polyp segmentation

    Deng-Ping Fan, Ge-Peng Ji, Tao Zhou, Geng Chen, Huazhu Fu, Jianbing Shen, and Ling Shao. Pranet: Parallel reverse at- tention network for polyp segmentation. In International con- ference on medical image computing and computer-assisted intervention, pages 263–273. Springer, 2020. 1

    2020

  4. [12]

    Concealed object detection

    Deng-Ping Fan, Ge-Peng Ji, Ming-Ming Cheng, and Ling Shao. Concealed object detection. IEEE Transactions on Pat- tern Analysis and Machine Intelligence, 44(10):6024–6042,

  5. [13]

    Semi- supervised camouflaged object detection from noisy data

    Yuanbin Fu, Jie Ying, Houlei Lv, and Xiaojie Guo. Semi- supervised camouflaged object detection from noisy data. In ACM International Conference on Multimedia, 2024. 6

    2024

  6. [14]

    Unsu- pervised representation learning by predicting image rotations

    Spyros Gidaris, Praveer Singh, and Nikos Komodakis. Unsu- pervised representation learning by predicting image rotations. In International Conference on Learning Representations ,

  7. [15]

    Obow: Online bag-of- visual-words generation for self-supervised learning

    Spyros Gidaris, Andrei Bursuc, Gilles Puy, Nikos Komodakis, Matthieu Cord, and Patrick Pérez. Obow: Online bag-of- visual-words generation for self-supervised learning. In The IEEE/CVF Computer Vision and Pattern Recognition Confer- ence, pages 6830–6840, 2021. 3

    2021

  8. [16]

    Context- aware saliency detection

    Stas Goferman, Lihi Zelnik-Manor, and Ayellet Tal. Context- aware saliency detection. IEEE transactions on pattern anal- ysis and machine intelligence, 34(10):1915–1926, 2011. 2

    1915

  9. [17]

    Generative adversarial nets

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. Advances in neural information processing systems, 27, 2014. 2

    2014

  10. [18]

    Bootstrap your own latent-a new approach to self-supervised learning

    Jean-Bastien Grill, Florian Strub, Florent Altché, Corentin Tallec, Pierre Richemond, Elena Buchatskaya, Carl Doer- sch, Bernardo Avila Pires, Zhaohan Guo, Mohammad Ghesh- laghi Azar, et al. Bootstrap your own latent-a new approach to self-supervised learning. Advances in neur...

    2020

  11. [19]

    Camouflaged object detection with feature decomposition and edge reconstruction

    Chunming He, Kai Li, Yachao Zhang, Longxiang Tang, Yu- lun Zhang, Zhenhua Guo, and Xiu Li. Camouflaged object detection with feature decomposition and edge reconstruction. In The IEEE/CVF Computer Vision and Pattern Recognition Conference, pages 22046–22055, 2023. 1

    2023

  12. [20]

    Momentum contrast for unsupervised visual repre- sentation learning

    Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross Girshick. Momentum contrast for unsupervised visual repre- sentation learning. In The IEEE/CVF Computer Vision and Pattern Recognition Conference, pages 9729–9738, 2020. 3

    2020

  13. [21]

    Masked autoencoders are scalable vision learners

    Kaiming He, Xinlei Chen, Saining Xie, Yanghao Li, Piotr Dollár, and Ross Girshick. Masked autoencoders are scalable vision learners. In The IEEE/CVF Computer Vision and Pattern Recognition Conference, pages 16000–16009, 2022. 3

    2022

  14. [22]

    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. In Association for the Advancement of Artificial Intelligence, pages 881–889, 2023. 6

    2023

  15. [23]

    Feature shrinkage pyra- mid for camouflaged object detection with transformers

    Zhou Huang, Hang Dai, Tian-Zhu Xiang, Shuo Wang, Huai- Xin Chen, Jie Qin, and Huan Xiong. Feature shrinkage pyra- mid for camouflaged object detection with transformers. In The IEEE/CVF Computer Vision and Pattern Recognition Conference, pages 5557–5566, 2023. 6

    2023

  16. [24]

    A model of saliency-based visual attention for rapid scene analysis

    Laurent Itti, Christof Koch, and Ernst Niebur. A model of saliency-based visual attention for rapid scene analysis. IEEE Transactions on pattern analysis and machine intelligence , 20(11):1254–1259, 1998. 2

    1998

  17. [25]

    Deep gradient learning for efficient camouflaged object detection

    Ge-Peng Ji, Deng-Ping Fan, Yu-Cheng Chou, Dengxin Dai, Alexander Liniger, and Luc Van Gool. Deep gradient learning for efficient camouflaged object detection. Machine Intelli- gence Research, 20:92–108, 2023. 3

    2023

  18. [26]

    Learning to predict where humans look

    Tilke Judd, Krista Ehinger, Frédo Durand, and Antonio Tor- ralba. Learning to predict where humans look. In The IEEE/CVF International Conference on Computer Vision , pages 2106–2113. IEEE, 2009. 2

    2009

  19. [27]

    Camoteacher: Dual-rotation consistency learning for semi-supervised camouflaged object detection

    Xunfa Lai, Zhiyu Yang, Jie Hu, Shengchuan Zhang, Liujuan Cao, Guannan Jiang, Zhiyu Wang, Songan Zhang, and Ron- grong Ji. Camoteacher: Dual-rotation consistency learning for semi-supervised camouflaged object detection. In European Conference on Computer Vision, 2024. 1, 6

    2024

  20. [28]

    Jiaying Lin, Xin Tan, Ke Xu, Lizhuang Ma, and Rynson W. H. Lau. Frequency-aware camouflaged object detection. ACM Trans. Multimedia Comput. Commun. Appl., 19(2), 2023. 3

    2023

  21. [29]

    Weakly supervised open-vocabulary object detection

    Jianghang Lin, Yunhang Shen, Bingquan Wang, Shaohui Lin, Ke Li, and Liujuan Cao. Weakly supervised open-vocabulary object detection. In Proceedings of the AAAI Conference on Artificial Intelligence, pages 3404–3412, 2024. 3

    2024

  22. [30]

    Camouflaged instance segmentation via explicit de-camouflaging

    Naisong Luo, Yuwen Pan, Rui Sun, Tianzhu Zhang, Zhiwei Xiong, and Feng Wu. Camouflaged instance segmentation via explicit de-camouflaging. In The IEEE/CVF Computer Vision and Pattern Recognition Conference, pages 17918–17927,

  23. [31]

    How to evaluate foreground maps? In The IEEE/CVF Computer Vision and Pattern Recognition Conference, pages 248–255,

    Ran Margolin, Lihi Zelnik-Manor, and Ayellet Tal. How to evaluate foreground maps? In The IEEE/CVF Computer Vision and Pattern Recognition Conference, pages 248–255,

  24. [32]

    Deep spectral methods: A surprisingly strong baseline for unsupervised semantic segmentation and local- ization

    Luke Melas-Kyriazi, Christian Rupprecht, Iro Laina, and An- drea Vedaldi. Deep spectral methods: A surprisingly strong baseline for unsupervised semantic segmentation and local- ization. In The IEEE/CVF Computer Vision and Pattern Recognition Conference, pages 8364–8375, 2022. 3, 6

    2022

  25. [33]

    Active teacher for semi-supervised object detection

    Peng Mi, Jianghang Lin, Yiyi Zhou, Yunhang Shen, Gen Luo, Xiaoshuai Sun, Liujuan Cao, Rongrong Fu, Qiang Xu, and Rongrong Ji. Active teacher for semi-supervised object detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 14482–14491,

  26. [34]

    Deepusps: Deep robust unsupervised saliency prediction via self-supervision

    Tam Nguyen, Maximilian Dax, Chaithanya Kumar Mummadi, Nhung Ngo, Thi Hoai Phuong Nguyen, Zhongyu Lou, and Thomas Brox. Deepusps: Deep robust unsupervised saliency prediction via self-supervision. Advances in Neural Informa- tion Processing Systems, 32, 2019. 2

    2019

  27. [35]

    Unsupervised learning of visual representations by solving jigsaw puzzles

    Mehdi Noroozi and Paolo Favaro. Unsupervised learning of visual representations by solving jigsaw puzzles. In European conference on computer vision, pages 69–84. Springer, 2016. 3

    2016

  28. [36]

    Maxime Oquab, Timothée Darcet, Theo Moutakanni, Huy V . V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Russell Howes, Po-Yao Huang, Hu Xu, Vasu Sharma, Shang-Wen Li, Wojciech Galuba, Mike Rabbat, Mido Assran, Nicolas...

  29. [37]

    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. In The IEEE/CVF Computer Vision and Pattern Recognition Conference, 2022. 2, 3, 5, 6

    2022

  30. [38]

    Zoomnext: A unified collaborative pyramid network for camouflaged object detection

    Youwei Pang, Xiaoqi Zhao, Tian-Zhu Xiang, Lihe Zhang, and Huchuan Lu. Zoomnext: A unified collaborative pyramid network for camouflaged object detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024. 1, 2, 3, 5

    2024

  31. [39]

    Saliency filters: Contrast based filtering for salient region detection

    Federico Perazzi, Philipp Krähenbühl, Yael Pritch, and Alexander Hornung. Saliency filters: Contrast based filtering for salient region detection. In The IEEE/CVF Computer Vision and Pattern Recognition Conference, pages 733–740. IEEE, 2012. 6

    2012

  32. [40]

    Improving language understanding by genera- tive pre-training

    Alec Radford. Improving language understanding by genera- tive pre-training. OpenAI blog, 2018. 3

    2018

  33. [41]

    Language models are unsuper- vised multitask learners

    Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever, et al. Language models are unsuper- vised multitask learners. OpenAI blog, 1(8):9, 2019. 3

    2019

  34. [42]

    Deep texture-aware features for camouflaged object detec- tion

    Jingjing Ren, Xiaowei Hu, Lei Zhu, Xuemiao Xu, Yangyang Xu, Weiming Wang, Zijun Deng, and Pheng-Ann Heng. Deep texture-aware features for camouflaged object detec- tion. IEEE Transactions on Circuits and Systems for Video Technology, 33(3):1157–1167, 2023. 3

    2023

  35. [43]

    Unsuper- vised salient object detection with spectral cluster voting

    Gyungin Shin, Samuel Albanie, and Weidi Xie. Unsuper- vised salient object detection with spectral cluster voting. In The IEEE/CVF Computer Vision and Pattern Recognition Conference, pages 3971–3980, 2022. 3, 6

    2022

  36. [44]

    Unsupervised object local- ization: Observing the background to discover objects

    Oriane Siméoni, Chloé Sekkat, Gilles Puy, Antonín V obeck`y, Éloi Zablocki, and Patrick Pérez. Unsupervised object local- ization: Observing the background to discover objects. In The IEEE/CVF Computer Vision and Pattern Recognition Conference, pages 3176–3186, 2023. 1, 2, 3, ...

    2023

  37. [45]

    Edge-aware mirror network for camouflaged object detection, 2023

    Dongyue Sun, Shiyao Jiang, and Lin Qi. Edge-aware mirror network for camouflaged object detection, 2023. 3

    2023

  38. [46]

    Bert4rec: Sequential recommendation with bidirectional encoder representations from transformer

    Fei Sun, Jun Liu, Jian Wu, Changhua Pei, Xiao Lin, Wenwu Ou, and Peng Jiang. Bert4rec: Sequential recommendation with bidirectional encoder representations from transformer. In Proceedings of the 28th ACM international conference on information and knowledge management, pages ...

  39. [47]

    Context-aware cross-level fusion network for camouflaged object detection

    Yujia Sun, Geng Chen, Tao Zhou, Yi Zhang, and Nian Liu. Context-aware cross-level fusion network for camouflaged object detection. In The International Joint Conference on Artificial Intelligence, pages 1025–1031, 2021. 6

    2021

  40. [48]

    Boundary-guided camouflaged object detection

    Yujia Sun, Shuo Wang, Chenglizhao Chen, and Tian-Zhu Xiang. Boundary-guided camouflaged object detection. In The International Joint Conference on Artificial Intelligence, pages 1335–1341, 2022. 6

    2022

  41. [49]

    Glconet: Learning multisource perception representation for camou- flaged object detection

    Yanguang Sun, Hanyu Xuan, Jian Yang, and Lei Luo. Glconet: Learning multisource perception representation for camou- flaged object detection. IEEE Transactions on Neural Net- works and Learning Systems, 2024. 1

    2024

  42. [50]

    Segmentation-based deep-learning approach for surface-defect detection

    Domen Tabernik, Samo Šela, Jure Skvar ˇc, and Danijel Skoˇcaj. Segmentation-based deep-learning approach for surface-defect detection. Journal of Intelligent Manufactur- ing, 31(3):759–776, 2020. 1

    2020

  43. [51]

    Ob- ject segmentation without labels with large-scale generative models

    Andrey V oynov, Stanislav Morozov, and Artem Babenko. Ob- ject segmentation without labels with large-scale generative models. In International Conference on Machine Learning, pages 10596–10606. PMLR, 2021. 6

    2021

  44. [52]

    Depth-aided camouflaged object detection

    Qingwei Wang, Jinyu Yang, Xiaosheng Yu, Fangyi Wang, Peng Chen, and Feng Zheng. Depth-aided camouflaged object detection. In Proceedings of the 31st ACM International Conference on Multimedia, page 3297–3306, 2023. 3

    2023

  45. [53]

    Self-supervised trans- formers for unsupervised object discovery using normalized cut

    Yangtao Wang, Xi Shen, Shell Xu Hu, Yuan Yuan, James L Crowley, and Dominique Vaufreydaz. Self-supervised trans- formers for unsupervised object discovery using normalized cut. In The IEEE/CVF Computer Vision and Pattern Recogni- tion Conference, pages 14543–14553, 2022. 3, 6

    2022

  46. [54]

    Geodesic saliency using background priors

    Yichen Wei, Fang Wen, Wangjiang Zhu, and Jian Sun. Geodesic saliency using background priors. In Computer Vision–ECCV 2012: 12th European Conference on Computer Vision, Florence, Italy, October 7-13, 2012, Proceedings, Part III 12, pages 29–42. Springer, 2012. 2

    2012

  47. [55]

    Vector-decomposed disentanglement for domain-invariant ob- ject detection

    Aming Wu, Rui Liu, Yahong Han, Linchao Zhu, and Yi Yang. Vector-decomposed disentanglement for domain-invariant ob- ject detection. The IEEE/CVF International Conference on Computer Vision, 2021. 5

    2021

  48. [56]

    Exploring depth contribution for camouflaged object detection, 2022

    Mochu Xiang, Jing Zhang, Yunqiu Lv, Aixuan Li, Yiran Zhong, and Yuchao Dai. Exploring depth contribution for camouflaged object detection, 2022. 3

    2022

  49. [57]

    Hierarchical saliency detection

    Qiong Yan, Li Xu, Jianping Shi, and Jiaya Jia. Hierarchical saliency detection. In The IEEE/CVF Computer Vision and Pattern Recognition Conference, pages 1155–1162, 2013. 2

    2013

  50. [58]

    Uncertainty-guided transformer reasoning for camouflaged object detection

    Fan Yang, Qiang Zhai, Xin Li, Rui Huang, Ao Luo, Hong Cheng, and Deng-Ping Fan. Uncertainty-guided transformer reasoning for camouflaged object detection. InThe IEEE/CVF International Conference on Computer Vision, pages 4146– 4155, 2021. 6

    2021

  51. [59]

    Hierarchical graph interaction transformer with dynamic token clustering for camouflaged object detection

    Siyuan Yao, Hao Sun, Tian-Zhu Xiang, Xiao Wang, and Xi- aochun Cao. Hierarchical graph interaction transformer with dynamic token clustering for camouflaged object detection. IEEE Transactions on Image Processing, 2024. 1

    2024

  52. [60]

    Camo- former: 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. Camo- former: Masked separable attention for camouflaged object detection. IEEE Transactions on Pattern Analysis and Ma- chine Intelligence, 2024. 1

    2024

  53. [61]

    Mutual graph learning for cam- ouflaged object detection

    Qiang Zhai, Xin Li, Fan Yang, Chenglizhao Chen, Hong Cheng, and Deng-Ping Fan. Mutual graph learning for cam- ouflaged object detection. In The IEEE/CVF Computer Vision and Pattern Recognition Conference, pages 12997–13007,

  54. [62]

    Deep unsupervised saliency detection: A multiple noisy labeling perspective

    Jing Zhang, Tong Zhang, Yuchao Dai, Mehrtash Harandi, and Richard Hartley. Deep unsupervised saliency detection: A multiple noisy labeling perspective. In The IEEE/CVF Computer Vision and Pattern Recognition Conference, pages 9029–9038, 2018. 2

    2018

  55. [63]

    Unsupervised camouflaged object segmentation as domain adaptation

    Yi Zhang and Chengyi Wu. Unsupervised camouflaged object segmentation as domain adaptation. In The IEEE/CVF Inter- national Conference on Computer Vision, pages 4334–4344,

  56. [64]

    Bilateral reference for high-resolution dichotomous image segmentation

    Peng Zheng, Dehong Gao, Deng-Ping Fan, Li Liu, Jorma Laaksonen, Wanli Ouyang, and Nicu Sebe. Bilateral reference for high-resolution dichotomous image segmentation. CAAI Artificial Intelligence Research, 3:9150038, 2024. 6

    2024

  57. [65]

    Detecting camouflaged object in frequency domain

    Yijie Zhong, Bo Li, Lv Tang, Senyun Kuang, Shuang Wu, and Shouhong Ding. Detecting camouflaged object in frequency domain. In The IEEE/CVF Computer Vision and Pattern Recognition Conference, pages 4494–4503, 2022. 3

    2022

  58. [66]

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

    Hongwei Zhu, Peng Li, Haoran Xie, Xu Yan, Dong Liang, Dapeng Chen, Mingqiang Wei, and Jing Qin. I can find you! boundary-guided separated attention network for camou- flaged object detection. In Association for the Advancement of Artificial Intelligence, 2022. 3

    2022

  59. [67]

    Inferring camouflaged objects by texture-aware interactive guidance network

    Jinchao Zhu, Xiaoyu Zhang, Shuo Zhang, and Junnan Liu. Inferring camouflaged objects by texture-aware interactive guidance network. In Association for the Advancement of Artificial Intelligence, 2021. 3

    2021

  60. [68]

    Saliency optimization from robust background detection

    Wangjiang Zhu, Shuang Liang, Yichen Wei, and Jian Sun. Saliency optimization from robust background detection. In The IEEE/CVF Computer Vision and Pattern Recognition Conference, pages 2814–2821, 2014. 2

    2014

  61. [69]

    Cubenet: X-shape connection for camouflaged object detection

    Mingchen Zhuge, Xiankai Lu, Yiyou Guo, Zhihua Cai, and Shuhan Chen. Cubenet: X-shape connection for camouflaged object detection. Pattern Recognition, 127:108644, 2022. 3

    2022

Pith tools

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