Pith. sign in

REVIEW 4 major objections 5 minor 60 references

CL-Polyp: A Contrastive Learning-Enhanced Network for Accurate Polyp Segmentation

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read CL-Polyp claims that adding a contrastive-learning branch to a DeepLabV3+ polyp segmentation network, together with two lightweight fusion modules, lifts segmentation accuracy beyond published state-of-the-art results.

desk verdict A plausible but poorly evidenced contrastive-learning polyp segmentation paper whose own generalization results contradict its central claim. read the letter →

arxiv 2507.07154 v2 pith:FCDXOXFG submitted 2025-07-09 cs.CV

Desheng Li , Chaoliang Liu , Zhiyong Xiao This is my paper
classification cs.CV
keywords polypsegmentationcontrastivelearningcolonoscopymedicalimageatrousspatialpyramidpoolingmomentumencodertripletlossDeepLabV3+
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

CL-Polyp is a polyp segmentation network that adds a self-supervised contrastive-learning task to the encoder of a DeepLabV3+-style baseline. The paper claims that contrasting positive and negative polyp images, selected by polyp number and size, improves the encoder's visual features without any additional annotations, and that two lightweight modules, MASPP for multi-scale feature fusion and CA for boundary reconstruction, further refine the output. On Kvasir-SEG the method reports a mean Dice of 0.918 and IoU of 0.864; on CVC-ClinicDB it reports Dice of 0.955 and IoU of 0.915. The authors position this as a route to the benefits of multi-task learning without task-specific labels or assumptions about task similarity.

What carries the argument

The load-bearing mechanism is the contrastive-learning branch attached to the ResNet50 encoder. A query feature $h$ taken from the image is compared, through a triplet loss, with a positive feature $h_p$ from an augmented view and $K=4$ negative features $h_n$ from images in different polyp number/size categories; the key features come from a momentum-updated encoder updated as $\theta_k = m\theta_k + (1-m)\theta_q$. Two decoder-side modules carry the rest of the design: MASPP, a modified atrous spatial pyramid pooling module that fuses the ASPP output with a $3\times3$-convolved identity branch through a squeeze-and-excitation module instead of plain channel concatenation, and the CA module, which merges low-level features $f_2$ with upsampled decoder features using 1x1 convolutions and concatenation before element-wise addition. Together these provide multi-scale feature fusion and boundary reconstruction.

What would settle it

Re-run each baseline in Tables 1-3 under the paper's stated splits and preprocessing, using the same evaluation code, and verify the reproduced numbers match the table. If PraNet's CVC-300 IoU cannot be reproduced at 9.797, the comparison is corrupted; if any baseline out-scores CL-Polyp under identical conditions, the claimed gains are refuted.

Watch

Extended reading notes

Core claim

The central discovery, stated on the paper's own terms, is that an auxiliary contrastive branch can regularize a segmentation encoder more effectively than the classification branches used in earlier multi-task polyp models. Positive samples are augmentations of the same image, while negative samples are images with different polyp number and size categories following the TGANet taxonomy. A momentum-updated encoder produces key features, and a triplet loss pulls the query feature toward its positive sample and away from negative samples while the segmentation branch is trained with a combined BCE and Dice loss. The full objective is $L_{\mathrm{Total}}=L_{\mathrm{Seg}}+\beta L_{\mathrm{CL}}$ with $\beta=0.5$, and the ablation study credits the contrastive branch with the largest single improvement. On the three unseen generalization datasets the model stays competitive but lands about two points below the best published results, which the paper attributes to contrastive training anchoring the model to the training distribution.

Load-bearing premise

The claimed lead over prior methods assumes that every baseline in Tables 1-3 was trained and evaluated under the same data splits and protocol as CL-Polyp (700/300 for Scenario I, PraNet's 90/10 split for Scenario II); the tables contain an impossible overlap score of 9.797 for PraNet on CVC-300, so this assumption is not established.

Editorial extensions

If this is right

  • On Kvasir-SEG with the 700/300 split, CL-Polyp reports mean Dice 0.918 and IoU 0.864, the best values among the compared methods on Dice, IoU, precision, and F2.
  • Under the PraNet 90/10 protocol, CL-Polyp reports Dice 0.955 and IoU 0.915 on CVC-ClinicDB, exceeding LDNet by 1.2 and 2.0 percentage points.
  • The ablation study shows the contrastive branch alone raises Dice by 1.8 and IoU by 2.2 over the MASPP+CA model, while MASPP and CA each contribute smaller gains over the DeepLabV3+ baseline.
  • On CVC-ColonDB, CVC-300, and ETIS, which are used only for testing, CL-Polyp remains competitive but does not reach the best published values, leaving a gap of about two points.

Reading between the lines

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

  • Editorial inference: because positives are augmentations and negatives are defined by polyp number/size categories, the same contrastive recipe could be dropped into other segmentation backbones, such as transformer encoders, as an auxiliary loss; this would test whether the gain comes from encoder regularization rather than the two proposed modules.
  • Editorial inference: the gap on the three cross-domain datasets suggests a testable hypothesis that the contrastive prior over training-set polyp categories reduces robustness to domain shift; an explicit domain-invariant contrastive objective could be evaluated on CVC-300 and ETIS.
  • Editorial inference: the odd values in the comparison tables, such as PraNet's CVC-300 IoU of 9.797, mean the claimed improvements over baselines should be treated as provisional until baselines are rerun under the same splits and preprocessing.
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 CL-Polyp, a polyp segmentation network built on a ResNet50 encoder and a DeepLabV3+-style decoder, augmented by an auxiliary contrastive learning branch. The contrastive branch uses a momentum-updated encoder, triplet loss, and positive/negative samples selected according to the TGANet polyp number/size classification. Two additional modules, MASPP and CA, are introduced for multi-scale fusion and boundary reconstruction. The method is evaluated on Kvasir-SEG, CVC-ClinicDB, CVC-ColonDB, CVC-300, and ETIS under two training scenarios, with further ablations on Kvasir-SEG.

Significance. If the reported results were reliable, the contribution of a contrastive branch that improves segmentation without additional manual annotations would be moderately useful for the polyp segmentation community. The paper provides explicit pseudo-code, a clear architecture figure, and ablation tables, which are useful for reproducibility in principle. However, the central claim of consistent state-of-the-art performance is contradicted by the paper's own results on three of the five datasets, there is an impossible baseline value in Table 3, and the self-supervised characterization of the contrastive branch is questionable because the sample categories are label-derived. No code, no statistical significance testing, and no sensitivity analysis for important hyperparameters are provided. As a result, the contribution is not currently demonstrated.

major comments (4)
  1. [Abstract; Section 4.4.2; Table 3] The abstract's claim that CL-Polyp 'consistently surpasses state-of-the-art methods' is contradicted by Table 3. On CVC-ColonDB, CL-Polyp (Dice/IoU 0.770/0.710) is below Polyp-PVT (0.808/0.727), BoxPolyp-Res2Net (0.820/0.741), and LDNet (0.784/0.706). On CVC-300, CL-Polyp (0.880/0.812) is below BoxPolyp-Res2Net (0.903/0.835) and Polyp-PVT (0.900/0.833). On ETIS, CL-Polyp (0.760/0.658) is below BoxPolyp-Res2Net (0.829/0.742) and Polyp-PVT (0.787/0.706). Section 4.4.2 explicitly concedes that CL-Polyp 'fails to reach the optimal index, and there is a gap of about 2% between our method and the other most advanced methods.' The stated global conclusion is therefore not supported by the reported experiments.
  2. [Table 3] The PraNet row reports a CVC-300 IoU of 9.797, which is impossible because IoU is bounded above by 1. This indicates a transcription or protocol error in the comparison table. Additionally, the paper does not state whether the baseline numbers in Tables 1-3 were produced by retraining under the same protocols (the 700/300 split in Scenario I and the PraNet 90/10 split in Scenario II) or were quoted from prior publications with different splits. Without this information, the claimed gains on Kvasir-SEG and CVC-ClinicDB in Table 2 cannot be considered reliable.
  3. [Section 3.1.1; Section 3.1.4] The contrastive branch is described as self-supervised and as not requiring additional annotations, but the selection of positive and negative samples relies on the TGANet categories by polyp number and size. These categories are derived from the ground-truth polyp masks, so the auxiliary task uses label information. The paper must either clarify how the categories are obtained without segmentation labels or revise the 'without needing additional annotations' claim.
  4. [Section 3.1.3; Algorithm 1] Equation (2) defines the triplet loss as a sum over N instances with one negative sample each, while the surrounding text and Algorithm 1 state that each image uses K negative samples. The notation should be aligned, for example by averaging or summing over K correctly. Moreover, the hyperparameters beta, K, alpha, and momentum m are fixed without any sensitivity analysis, so the contribution of the contrastive branch to the reported improvements is not isolated.
minor comments (5)
  1. [Throughout] There are numerous typos and ungrammatical sentences: 'Resluts' in Section 4.4.2, 'Iou' in Table 2, 'Kvaisr' in Section 4.1, and wording such as 'some things could still be improved so that we can achieve good results' in Section 3.2. These should be corrected.
  2. [Section 3.2.2] The CA module is defined by the self-referential expression zi = Ai + Conv(cat(Bi, zi)), which is ambiguous because zi appears on both sides. A clear forward-computation description or a diagram with variable names is needed.
  3. [Section 1; Section 2] The Introduction and Related Work contain many references to topics that are unrelated to polyp segmentation, including hyperspectral NMF, parallel genetic algorithms, MR image reconstruction, food recognition, and EEG analysis. These should be removed or moved to a focused discussion, as they distract from the paper's contribution.
  4. [Algorithm 1] The pseudo-code line 'for x, GT, K ∗ xn in loader do' is not valid PyTorch syntax and is unclear about how K negative samples are obtained and batched. It should be rewritten as a concrete data-loading loop.
  5. [Section 4.2] No code, trained models, or detailed hyperparameter search are provided. Given the comparison-protocol concerns in the major comments, releasing code or at least a precise description of the training and evaluation pipeline is necessary for the results to be reproducible.

Circularity Check

2 steps flagged · score 4.0 of 10

CL-Polyp's reported gains are empirically grounded but compromised: the final augmentation is chosen on the same Kvasir-SEG test split whose score is then reported, and the 'self-supervised' contrastive branch actually derives its positive/negative categories from ground-truth labels.

  1. fitted input called prediction [Section 4.4.3, Table 5; final results in Tables 1-2]
    "Data enhancement methods are essential for learning data invariance by contrast learning, so we explored the influence of different data enhancement methods on our methods. Table 5 shows the experimental results... The ablation results show that adding only Gaussian blur to the Base data augmentation yields the best results."

    The augmentation set is selected by running an ablation on the Kvasir-SEG 700/300 split described in Experiment Scenario III as 'divided in the same way as that of Experiment Scenario I'. The same split is then used to report CL-Polyp's headline Kvasir-SEG performance (0.918 Dice / 0.864 IoU), including the claimed 0.011 IoU gain over GLFRNet. The reported test-set number is therefore the selection criterion for the augmentation, not an independent prediction of it; the design is fit to the very test set on which the improvement is claimed.

  2. other [Section 3.1.1, Section 1]
    "According to the classification method of data in TGANet[4], polyps are divided into one or more according to the number of polyps and the size. ... This categorization facilitates the selection of positive and negative samples. ... This self-supervised strategy improves visual representation without needing additional annotations."

    The contrastive branch's positive and negative samples are defined by polyp number and size classes, which are properties read from the ground-truth masks used for the segmentation task. Calling the branch 'self-supervised' and claiming it needs no additional annotations is inconsistent: the auxiliary signal is label-derived. The segmentation output is still supervised on held-out masks, so this issue undercuts the self-supervision claim rather than forcing the segmentation result itself.

full rationale

The paper contains no derivation chain in the mathematical sense; it is an empirical architecture-and-training study. The central claim that CL-Polyp 'consistently surpasses state-of-the-art methods' is contradicted by the authors' own Table 3 on CVC-ColonDB, CVC-300, and ETIS, where several baselines are stronger, and Table 3 also contains an impossible PraNet IoU value (9.797). Those are correctness/comparability problems, not circularity. The self-citations in the introduction and related work are numerous but not load-bearing; none is invoked to forbid alternatives or to justify a uniqueness claim. The two genuine circularity-adjacent issues are (1) the final augmentation configuration is selected by ablation on the same Kvasir-SEG test split whose improvement is then reported, which makes that specific reported gain partly a selection artifact, and (2) the so-called self-supervised contrastive branch derives its sample categories from ground-truth polyp properties, weakening the claim of annotation-free representation learning. Because the segmentation evaluation still uses held-out masks and external baselines, the overall score is moderate rather than high.

Assumptions & free parameters 5 free parameters · 3 assumptions · 0 invented entities

The reported performance depends on hyperparameters chosen by hand, external category labels from TGANet that are not part of the public datasets, and the implicit assumption that baseline numbers are comparable. The augmentation selection on the test set further weakens the independent grounding of the claimed gains.

free parameters (5)
  • beta (contrastive loss weight) = 0.5
    Set by hand in Eq. (7); no sensitivity study reported.
  • K (number of negative samples) = 4
    Set to 4 in Section 3.1.3; no ablation on K.
  • triplet margin alpha = not reported
    Margin in triplet loss Eq. (2); value not given in the paper.
  • momentum parameter m = not reported
    Momentum update in Eq. (1); value not given.
  • data augmentation set = Gaussian blur selected as best
    The augmentation combination was chosen by comparing test-set Dice and IoU in Table 5.
assumptions (3)
  • domain assumption Polyp count and size categories from TGANet are available for all training images and are sufficient for selecting contrastive negatives.
    Used in Section 3.1.1 to define positive and negative samples, yet the standard datasets do not ship these labels; the paper does not explain how they were obtained.
  • domain assumption Reported baseline numbers in Tables 1-3 are produced under the same training protocol and are directly comparable.
    The paper does not state whether baselines were retrained; Table 3 contains an impossible value.
  • standard math Held-out test sets measure generalization.
    Standard evaluation practice, but the paper uses test performance to select data augmentation, weakening this assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CL-Polyp: A Contrastive Learning-Enhanced Network for Accurate Polyp Segmentation." pith.science (2026). https://pith.science/paper/FCDXOXFG

@misc{pith2026250707154,
  author       = {Pith},
  title        = {Pith review of: CL-Polyp: A Contrastive Learning-Enhanced Network for Accurate Polyp Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FCDXOXFG}},
  note         = {Machine review of arXiv:2507.07154}
}
read the original abstract

Accurate segmentation of polyps from colonoscopy images is crucial for the early diagnosis and treatment of colorectal cancer. Most existing deep learning-based polyp segmentation methods adopt an Encoder-Decoder architecture, and some utilize multi-task frameworks that incorporate auxiliary tasks like classification to improve segmentation. However, these methods often need more labeled data and depend on task similarity, potentially limiting generalizability. To address these challenges, we propose CL-Polyp, a contrastive learning-enhanced polyp segmentation network. Our method uses contrastive learning to enhance the encoder's extraction of discriminative features by contrasting positive and negative sample pairs from polyp images. This self-supervised strategy improves visual representation without needing additional annotations. We also introduce two efficient, lightweight modules: the Modified Atrous Spatial Pyramid Pooling (MASPP) module for improved multi-scale feature fusion, and the Channel Concatenate and Element Add (CA) module to merge low-level and upsampled features for {enhanced} boundary reconstruction. Extensive experiments on five benchmark datasets-Kvasir-SEG, CVC-ClinicDB, CVC-ColonDB, CVC-300, and ETIS-show that CL-Polyp consistently surpasses state-of-the-art methods. Specifically, it enhances the IoU metric by 0.011 and 0.020 on the Kvasir-SEG and CVC-ClinicDB datasets, respectively, demonstrating its effectiveness in clinical polyp segmentation.

Figures

Figures reproduced from arXiv: 2507.07154 by the authors.

Figure 1
Figure 1. Polyp images from the Kvasir-SEG dataset[ [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Encoder-decoder structure variations for polyp segmentation models: (a) Balanced encoder and decoder; (b) [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Architecture of the ResNet50-based segmentation model with contrastive learning: Feature maps extracted by [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Segmentation model architecture: (a) Overall structure with ResNet50 encoder, MASPP, and CA modules in [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Qualitative comparison of segmentation results: Proposed method versus six state-of-the-art methods on [PITH_FULL_IMAGE:figures/full_fig_p012_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

60 extracted references · 43 canonical work pages

  1. [1]

    Siegel, Kimberly D

    Rebecca L. Siegel, Kimberly D. Miller, Ann Goding Sauer, Stacey A. Fedewa, Lynn F. Butterly, Joseph C. Anderson, Andrea Cercek, Robert A. Smith, and Ahmedin Jemal. Colorectal cancer statistics, 2020. CA: A Cancer Journal for Clinicians, 70:145–164, 5 2020

    work page 2020

  2. [2]

    Clustering sparse swarm decomposition for automated recognition of upper limb movements from non-homogeneous cross-channel eeg signals

    Shailesh Vitthalrao Bhalerao and Ram Bilas Pachori. Clustering sparse swarm decomposition for automated recognition of upper limb movements from non-homogeneous cross-channel eeg signals. IEEE Sensors Letters, 2023

    work page 2023

  3. [3]

    Smedsrud, Michael A

    Debesh Jha, Pia H. Smedsrud, Michael A. Riegler, Pål Halvorsen, Thomas de Lange, Dag Johansen, and Håvard D. Johansen. Kvasir-seg: A segmented polyp dataset. In Yong Man Ro, Wen-Huang Cheng, Junmo Kim, Wei-Ta Chu, Peng Cui, Jung-Woo Choi, Min-Chun Hu, and Wesley De Neve, editors, MultiMedia Modeling - 26th International Conference, MMM 2020, Daejeon, Sout...

    work page 2020

  4. [4]

    Tganet: Text-guided attention for improved polyp segmentation

    Nikhil Kumar Tomar, Debesh Jha, Ulas Bagci, and Sharib Ali. Tganet: Text-guided attention for improved polyp segmentation. In Linwei Wang, Qi Dou, P. Thomas Fletcher, Stefanie Speidel, and Shuo Li, editors, Medical Image Computing and Computer Assisted Intervention - MICCAI 2022 - 25th International Conference, Singapore, September 18-22, 2022, Proceeding...

    work page 2022

  5. [5]

    U-net: Convolutional networks for biomedical image segmentation

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In Nassir Navab, Joachim Hornegger, William M. Wells III, and Alejandro F. Frangi, editors, Medical Image Computing and Computer-Assisted Intervention - MICCAI 2015 - 18th International Conference Munich, Germany, October 5 - 9, 2015, Proce...

    work page 2015

  6. [6]

    Unet++: Redesigning skip connections to exploit multiscale features in image segmentation

    Zongwei Zhou, Md Mahfuzur Rahman Siddiquee, Nima Tajbakhsh, and Jianming Liang. Unet++: Redesigning skip connections to exploit multiscale features in image segmentation. IEEE Trans. Medical Imaging, 39(6):1856– 1867, 2020

    work page 2020

  7. [7]

    Smedsrud, Michael A

    Debesh Jha, Pia H. Smedsrud, Michael A. Riegler, Dag Johansen, Thomas de Lange, Pål Halvorsen, and Håvard D. Johansen. Resunet++: An advanced architecture for medical image segmentation. In IEEE International Symposium on Multimedia, ISM 2019, San Diego, CA, USA, December 9-11, 2019, pages 225–230. IEEE, 2019

    work page 2019

  8. [8]

    Constrained nonnegative matrix factorization and hyperspectral image dimensionality reduction

    Zhiyong Xiao and Salah Bourennane. Constrained nonnegative matrix factorization and hyperspectral image dimensionality reduction. REMOTE SENSING LETTERS, 5(1):46–54, 2014

    work page 2014

Show all 60 references
  1. [9]

    A spectral–spatial similarity-based method and its application to hyperspectral image classification

    Zhiyong) Xiao. A spectral–spatial similarity-based method and its application to hyperspectral image classification. Remote Sensing Letters, pages 122–130, 2014

    2014

  2. [10]

    Highly scalable parallel genetic algorithm on sunway many-core processors

    Zhiyong Xiao, Xu Liu, Jingheng Xu, Qingxiao Sun, and Lin Gan. Highly scalable parallel genetic algorithm on sunway many-core processors. Future Generation Computer Systems, pages 679–691, 2021

    2021

  3. [11]

    Shankaranarayana, Keerthi Ram, Jayaraj Joseph, and Mohanasankar Sivaprakasam

    Balamurali Murugesan, Kaushik Sarveswaran, Sharath M. Shankaranarayana, Keerthi Ram, Jayaraj Joseph, and Mohanasankar Sivaprakasam. Psi-net: Shape and boundary aware joint multi-task deep network for medical image segmentation. In 41st Annual International Conference of the IE...

    2019

  4. [12]

    Pranet: Parallel reverse attention network for polyp segmentation

    Deng-Ping Fan, Ge-Peng Ji, Tao Zhou, Geng Chen, Huazhu Fu, Jianbing Shen, and Ling Shao. Pranet: Parallel reverse attention network for polyp segmentation. In Anne L. Martel, Purang Abolmaesumi, Danail Stoyanov, Diana Mateus, Maria A. Zuluaga, S. Kevin Zhou, Daniel Racoceanu, ...

    2020

  5. [13]

    Kevin Zhou, and Shuguang Cui

    Jun Wei, Yiwen Hu, Ruimao Zhang, Zhen Li, S. Kevin Zhou, and Shuguang Cui. Shallow attention network for polyp segmentation. In Marleen de Bruijne, Philippe C. Cattin, Stéphane Cotin, Nicolas Padoy, Stefanie Speidel, 13 CL-Polyp: Contrastive Learning for Accurate Polyp Segment...

    2021

  6. [14]

    Lesion-aware dynamic kernel for polyp segmentation

    Ruifei Zhang, Peiwen Lai, Xiang Wan, De-Jun Fan, Feng Gao, Xiao-Jian Wu, and Guanbin Li. Lesion-aware dynamic kernel for polyp segmentation. In Linwei Wang, Qi Dou, P. Thomas Fletcher, Stefanie Speidel, and Shuo Li, editors, Medical Image Computing and Computer Assisted Interv...

    2022

  7. [15]

    Encoder-decoder with atrous separable convolution for semantic image segmentation

    Liang-Chieh Chen, Yukun Zhu, George Papandreou, Florian Schroff, and Hartwig Adam. Encoder-decoder with atrous separable convolution for semantic image segmentation. In Vittorio Ferrari, Martial Hebert, Cristian Sminchisescu, and Yair Weiss, editors, Computer Vision - ECCV 201...

    2018

  8. [16]

    Transfuse: Fusing transformers and cnns for medical image segmentation

    Yundong Zhang, Huiye Liu, and Qiang Hu. Transfuse: Fusing transformers and cnns for medical image segmentation. In Marleen de Bruijne, Philippe C. Cattin, Stéphane Cotin, Nicolas Padoy, Stefanie Speidel, Yefeng Zheng, and Caroline Essert, editors, Medical Image Computing and C...

    2021

  9. [17]

    Polyp-pvt: Polyp segmentation with pyramid vision transformers

    Bo Dong, Wenhai Wang, Deng-Ping Fan, Jinpeng Li, Huazhu Fu, and Ling Shao. Polyp-pvt: Polyp segmentation with pyramid vision transformers. CoRR, abs/2108.06932, 2021

    2021 arXiv

  10. [18]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In 2016 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016, Las Vegas, NV , USA, June 27-30, 2016, pages 770–778. IEEE Computer Society, 2016

    2016

  11. [19]

    Pyramid vision transformer: A versatile backbone for dense prediction without convolutions

    Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, and Ling Shao. Pyramid vision transformer: A versatile backbone for dense prediction without convolutions. In 2021 IEEE/CVF International Conference on Computer Vision, ICCV 2021, Montr...

    2021

  12. [20]

    High accuracy food image classification via vision transformer with data augmentation and feature augmentation

    Xinle Gao, Zhiyong Xiao, and Zhaohong Deng. High accuracy food image classification via vision transformer with data augmentation and feature augmentation. Journal of Food Engineering, 365:111833, 2024

    2024

  13. [21]

    Fine grained food image recognition based on swin transformer

    Zhiyong Xiao, Guang Diao, and Zhaohong Deng. Fine grained food image recognition based on swin transformer. Journal of Food Engineering, 380:112134, 2024

    2024

  14. [22]

    Foodcswin: A high-accuracy food image recognition model for dietary assessment

    Zhiyong Xiao, Ruke Ling, and Zhaohong Deng. Foodcswin: A high-accuracy food image recognition model for dietary assessment. Journal of Food Composition and Analysis, 139:107110, 2025

    2025

  15. [23]

    Fgfoodnet: Ingredient-perceived fine-grained food recognition for dietary monitoring

    Zhiyong Xiao, Yida Sun, and Zhaohong Deng. Fgfoodnet: Ingredient-perceived fine-grained food recognition for dietary monitoring. JOURNAL OF FOOD MEASUREMENT AND CHARACTERIZATION, 2025 JUN 28 2025

    2025

  16. [24]

    Fine-grained crop pest classification based on multi-scale feature fusion and mixed attention mechanisms

    Yiheng Qian, Zhiyong Xiao, and Zhaohong Deng. Fine-grained crop pest classification based on multi-scale feature fusion and mixed attention mechanisms. Frontiers in Plant Science, page 1500571, 2025

    2025

  17. [25]

    Swin attention augmented residual network: a fine-grained pest image recognition method

    Xiang Wang, Zhiyong Xiao, and Zhaohong Deng. Swin attention augmented residual network: a fine-grained pest image recognition method. Frontiers in plant science, page 1619551, 2025

    2025

  18. [26]

    Essdm: An enhanced sparse swarm decomposition method and its application in multi-class motor imagery–based eeg-bci system

    Shailesh Bhalerao and Ram Bilas Pachori. Essdm: An enhanced sparse swarm decomposition method and its application in multi-class motor imagery–based eeg-bci system. Authorea Preprints, 2023

    2023

  19. [27]

    Fcp-net: A feature-compression-pyramid network guided by game-theoretic interactions for medical image segmentation

    Yexin Liu, Jian Zhou, Lizhu Liu, Zhengjia Zhan, Yueqiang Hu, Yongqing Fu, and Huigao Duan. Fcp-net: A feature-compression-pyramid network guided by game-theoretic interactions for medical image segmentation. IEEE Trans. Medical Imaging, 41(6):1482–1496, 2022

    2022

  20. [28]

    Detection of tumor in liver using image segmentation and registration technique

    Priyanka Kumar and Shailesh Bhalerao. Detection of tumor in liver using image segmentation and registration technique. Journal of Electronics and Communication Engineering, 9(2):110–115, 2014

    2014

  21. [29]

    Imagined speech-eeg detection using multivariate swarm sparse decomposition-based joint time-frequency analysis for intuitive bci

    Shailesh Vitthalrao Bhalerao and Ram Bilas Pachori. Imagined speech-eeg detection using multivariate swarm sparse decomposition-based joint time-frequency analysis for intuitive bci. Authorea Preprints, 2024

    2024

  22. [30]

    Application of improved convolutional neural network in medical image segmentation

    Chen Liu, Zhiyong Xiao, and Nianmao Du. Application of improved convolutional neural network in medical image segmentation. Journal of Frontiers of Computer Science and Technology, pages 1593–1603, 2019

    2019

  23. [31]

    Application of improved convolutional neural network in lung image segmentation

    Baoxin Qian, Zhiyong Xiao, and Wei Song. Application of improved convolutional neural network in lung image segmentation. Journal of Frontiers of Computer Science and Technology, pages 1358–1367, 2020. 14 CL-Polyp: Contrastive Learning for Accurate Polyp Segmentation

    2020

  24. [32]

    3d u-net applied to simple attention module for head and neck tumor segmentation in pet and ct images

    Tao Liu, Yixin Su, Jiabao Zhang, Tianqi Wei, and Zhiyong Xiao. 3d u-net applied to simple attention module for head and neck tumor segmentation in pet and ct images. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Not...

    2022

  25. [33]

    Sr-net: A sequence offset fusion net and refine net for undersampled multislice mr image reconstruction

    Zhiyong Xiao, Nianmao Du, Jianjun Liu, and Weidong Zhang. Sr-net: A sequence offset fusion net and refine net for undersampled multislice mr image reconstruction. COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE, 202, 2021

    2021

  26. [34]

    Multi-view hierarchical split network for brain tumor segmentation

    Zhiyong Xiao, Kanghui He, Jianjun Liu, and Weidong Zhang. Multi-view hierarchical split network for brain tumor segmentation. Biomedical Signal Processing and Control, 69, 2021

    2021

  27. [35]

    Rmmlp:rolling mlp and matrix decomposition for skin lesion segmentation

    Chao Ji, Zhaohong Deng, Yan Ding, Fengsheng Zhou, and Zhiyong Xiao. Rmmlp:rolling mlp and matrix decomposition for skin lesion segmentation. Biomedical Signal Processing and Control, 84:104825, 2023

    2023

  28. [36]

    Efficient combination of cnn and transformer for dual-teacher uncertainty-guided semi-supervised medical image segmentation

    Zhiyong Xiao, Yixin Su, Zhaohong Deng, and Weidong Zhang. Efficient combination of cnn and transformer for dual-teacher uncertainty-guided semi-supervised medical image segmentation. Computer Methods and Programs in Biomedicine, 226:107099, 2022

    2022

  29. [37]

    Light3dhs: A lightweight 3d hippocampus segmentation method using multiscale convolution attention and vision transformer

    Zhiyong Xiao, Yuhong Zhang, Zhaohong Deng, and Fei Liu. Light3dhs: A lightweight 3d hippocampus segmentation method using multiscale convolution attention and vision transformer. NeuroImage, 292:120608, 2024

    2024

  30. [38]

    Fully convolutional networks for semantic segmentation

    Jonathan Long, Evan Shelhamer, and Trevor Darrell. Fully convolutional networks for semantic segmentation. In IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2015, Boston, MA, USA, June 7-12, 2015, pages 3431–3440. IEEE Computer Society, 2015

    2015

  31. [39]

    Squeeze-and-excitation networks

    Jie Hu, Li Shen, and Gang Sun. Squeeze-and-excitation networks. In 2018 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2018, Salt Lake City, UT, USA, June 18-22, 2018, pages 7132–7141. Computer Vision Foundation / IEEE Computer Society, 2018

    2018

  32. [40]

    Gomez, Lukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Isabelle Guyon, Ulrike von Luxburg, Samy Bengio, Hanna M. Wallach, Rob Fergus, S. V . N. Vishwanathan, and Roman Garnett,...

    2017

  33. [41]

    Hardnet-mseg: A simple encoder-decoder polyp segmentation neural network that achieves over 0.9 mean dice and 86 FPS

    Chien-Hsiang Huang, Hung-Yu Wu, and Youn-Long Lin. Hardnet-mseg: A simple encoder-decoder polyp segmentation neural network that achieves over 0.9 mean dice and 86 FPS. CoRR, abs/2101.07172, 2021

    2021 arXiv

  34. [42]

    Automatic polyp segmentation via multi-scale subtraction network

    Xiaoqi Zhao, Lihe Zhang, and Huchuan Lu. Automatic polyp segmentation via multi-scale subtraction network. In Marleen de Bruijne, Philippe C. Cattin, Stéphane Cotin, Nicolas Padoy, Stefanie Speidel, Yefeng Zheng, and Caroline Essert, editors, Medical Image Computing and Comput...

    2021

  35. [43]

    Ange Lou, Shuyue Guan, and Murray H. Loew. Caranet: Context axial reverse attention network for segmentation of small medical objects. CoRR, abs/2108.07368, 2021

    2021 arXiv

  36. [44]

    Global and local feature reconstruction for medical image segmentation

    Jiahuan Song, Xinjian Chen, Qianlong Zhu, Fei Shi, Dehui Xiang, Zhongyue Chen, Ying Fan, Lingjiao Pan, and Weifang Zhu. Global and local feature reconstruction for medical image segmentation. IEEE Trans. Medical Imaging, 41(9):2273–2284, 2022

    2022

  37. [45]

    Shanghua Gao, Ming-Ming Cheng, Kai Zhao, Xin-Yu Zhang, Ming-Hsuan Yang, and Philip H. S. Torr. Res2net: A new multi-scale backbone architecture. IEEE Trans. Pattern Anal. Mach. Intell., 43(2):652–662, 2021

    2021

  38. [46]

    Kevin Zhou, and Zhen Li

    Jun Wei, Yiwen Hu, Guanbin Li, Shuguang Cui, S. Kevin Zhou, and Zhen Li. Boxpolyp: Boost generalized polyp segmentation using extra coarse bounding box annotations. In Linwei Wang, Qi Dou, P. Thomas Fletcher, Stefanie Speidel, and Shuo Li, editors, Medical Image Computing and ...

    2022

  39. [47]

    Semi-supervised ct image segmentation via contrastive learning based on entropy constraints

    Zhiyong Xiao, Hao Sun, and Fei Liu. Semi-supervised ct image segmentation via contrastive learning based on entropy constraints. Biomedical Engineering Letters, pages 1023–1035, 2024

    2024

  40. [48]

    Frcnet: Feature refining and context-guided network for efficient polyp segmentation

    Liantao Shi, Yufeng Wang, Zhengguo Li, and Wen Qiumiao. Frcnet: Feature refining and context-guided network for efficient polyp segmentation. Frontiers in Bioengineering and Biotechnology, 10, 6 2022

    2022

  41. [49]

    Multi-task learning for segmentation and classification of tumors in 3d automated breast ultrasound images

    Yue Zhou, Houjin Chen, Yanfeng Li, Qin Liu, Xuanang Xu, Shu Wang, Pew-Thian Yap, and Dinggang Shen. Multi-task learning for segmentation and classification of tumors in 3d automated breast ultrasound images. Medical Image Anal., 70:101918, 2021. 15 CL-Polyp: Contrastive Learni...

    2021

  42. [50]

    Girshick

    Kaiming He, Haoqi Fan, Yuxin Wu, Saining Xie, and Ross B. Girshick. Momentum contrast for unsupervised visual representation learning. In 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2020, Seattle, WA, USA, June 13-19, 2020, pages 9726–9735. Comput...

    2020

  43. [51]

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey E. Hinton. A simple framework for contrastive learning of visual representations. In Proceedings of the 37th International Conference on Machine Learning, ICML 2020, 13-18 July 2020, Virtual Event, volume 119 of Procee...

    2020

  44. [52]

    Exploring simple siamese representation learning

    Xinlei Chen and Kaiming He. Exploring simple siamese representation learning. InIEEE Conference on Computer Vision and Pattern Recognition, CVPR 2021, virtual, June 19-25, 2021, pages 15750–15758. Computer Vision Foundation / IEEE, 2021

    2021

  45. [53]

    Facenet: A unified embedding for face recognition and clustering

    Florian Schroff, Dmitry Kalenichenko, and James Philbin. Facenet: A unified embedding for face recognition and clustering. In IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2015, Boston, MA, USA, June 7-12, 2015, pages 815–823. IEEE Computer Society, 2015

    2015

  46. [54]

    Automated classification of cognitive visual objects using multivariate swarm sparse decomposition from multichannel EEG-MEG signals

    Shailesh Vitthalrao Bhalerao and Ram Bilas Pachori. Automated classification of cognitive visual objects using multivariate swarm sparse decomposition from multichannel EEG-MEG signals. IEEE Trans. Hum. Mach. Syst., 54(4):455–464, 2024

    2024

  47. [55]

    Sparse spectrum based swarm decomposition for robust nonstationary signal analysis with application to sleep apnea detection from EEG

    Shailesh Vitthalrao Bhalerao and Ram Bilas Pachori. Sparse spectrum based swarm decomposition for robust nonstationary signal analysis with application to sleep apnea detection from EEG. Biomed. Signal Process. Control., 77:103792, 2022

    2022

  48. [56]

    Jaeger, Simon Kohl, Jakob Wasserthal, Gregor Köhler, Tobias Norajitra, Sebastian J

    Fabian Isensee, Jens Petersen, André Klein, David Zimmerer, Paul F. Jaeger, Simon Kohl, Jakob Wasserthal, Gregor Köhler, Tobias Norajitra, Sebastian J. Wirkert, and Klaus H. Maier-Hein. nnu-net: Self-adapting framework for u-net-based medical image segmentation. CoRR, abs/1809...

    2018 arXiv

  49. [57]

    WM-DOV A maps for accurate polyp highlighting in colonoscopy: Validation vs

    Jorge Bernal, Francisco Javier Sánchez, Gloria Fernández-Esparrach, Debora Gil, Cristina Rodríguez de Miguel, and Fernando Vilariño. WM-DOV A maps for accurate polyp highlighting in colonoscopy: Validation vs. saliency maps from physicians. Comput. Medical Imaging Graph., 43:9...

    2015

  50. [58]

    Gurudu, and Jianming Liang

    Nima Tajbakhsh, Suryakanth R. Gurudu, and Jianming Liang. Automated polyp detection in colonoscopy videos using shape and context information. IEEE Trans. Medical Imaging, 35(2):630–644, 2016

    2016

  51. [59]

    López, Adriana Romero, Michal Drozdzal, and Aaron C

    David Vázquez, Jorge Bernal, Francisco Javier Sánchez, Gloria Fernández-Esparrach, Antonio M. López, Adriana Romero, Michal Drozdzal, and Aaron C. Courville. A benchmark for endoluminal scene segmentation of colonoscopy images. CoRR, abs/1612.00799, 2016

    2016 arXiv

  52. [60]

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

    Juan Silva, Aymeric Histace, Olivier Romain, Xavier Dray, and Bertrand Granado. Toward embedded detection of polyps in WCE images for early diagnosis of colorectal cancer. Int. J. Comput. Assist. Radiol. Surg., 9(2):283–293, 2014. 16

    2014

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.