Back

PraNet: Parallel Reverse Attention Network for Polyp Segmentation

Deng-Ping FanGe-Peng JiTao ZhouGeng ChenHuazhu FuJianbing ShenLing Shao

article2020International Conference on Medical Image Computing and Computer-Assisted Intervention1,918 citationsMICCAI 2025 Young Scientist Publication Impact Award

Introduces PraNet, an efficient deep network that resolves vague lesion boundaries in colonoscopy images by coupling high-level context aggregation with reverse attention modules to achieve accurate, real-time colorectal polyp segmentation.

Listen

Colorectal cancer is the third most common cancer globally, making early screening and lesion removal a vital public health priority. While colonoscopy is standard for detecting precancerous polyps, accurate automatic polyp segmentation remains difficult. Polyps exhibit significant variations in color, size, and texture, and their boundaries blend into surrounding mucosa, leading to missed detections and inaccurate boundary outlines.

The article evaluates a deep neural network named PraNet (Parallel Reverse Attention Network) to demonstrate how combining high-level contextual area mapping with reverse attention boundary refinement can achieve accurate, real-time polyp segmentation.

The approach mirrors clinical practice by first predicting a coarse polyp area and then progressively refining the boundary. A parallel partial decoder aggregates high-level semantic features to produce a global guidance map, while recurrent reverse attention modules subtract estimated foreground regions from side-output features to capture subtle boundary cues. Evaluated across five benchmark colonoscopy datasets (ETIS, CVC-ClinicDB, CVC-ColonDB, EndoScene, and Kvasir), the model was trained in an end-to-end setting with weighted intersection-over-union and binary cross-entropy loss functions.

The evaluation produced four key findings. First, PraNet demonstrated superior learning accuracy, achieving an 89.8% to 89.9% mean Dice score on seen datasets, outperforming established architectures like U-Net, U-Net++, and selective feature aggregation networks by more than 7%. Second, the architecture showed strong generalization across unseen datasets, achieving a 70.9% mean Dice score on CVC-ColonDB and 62.8% on ETIS, where baseline methods dropped to 29.7%–51.2%. Third, PraNet achieved real-time inference at approximately 50 frames per second on standard hardware. Fourth, training was exceptionally efficient, converging in only 20 epochs (about 30 minutes) compared to over 20 hours required by competing edge-aggregation models.

These findings indicate that PraNet can enhance computer-aided colonoscopy tools by providing real-time, highly accurate guidance during live procedures without requiring complex pre- or post-processing. Because it effectively separates subtle polyp boundaries without overfitting, it reduces computational costs and lowers clinical risk associated with missed polyps. These results also show that using reverse attention to implicitly refine boundaries outperforms heavy, explicit edge-supervision models in both speed and robustness.

For operational deployment, stakeholders should evaluate integrating PraNet into real-time clinical video streams and validate performance in live procedural pilots. Researchers should also extend this reverse attention framework to other complex medical imaging tasks, such as lung infection segmentation.

Confidence in these findings is strong across the evaluated benchmarks; however, limitations remain regarding performance degradation on highly constrained datasets (such as the 62.8% Dice score on the small ETIS dataset), meaning additional diverse clinical data is recommended before widespread adoption.

Inception Institute of AIMohamed bin Zayed University of Artificial IntelligenceWuhan Universitysegmentation methodshigh resolution imageParallel Reverse Attention Networkpolyp segmentationbiomedical image segmentation
arXiv: 2006.11392DengPingFan/PraNet
Cover for PraNet: Parallel Reverse Attention Network for Polyp Segmentation

Abstract

Colonoscopy is an effective technique for detecting colorectal polyps, which are highly related to colorectal cancer. In clinical practice, segmenting polyps from colonoscopy images is of great importance since it provides valuable information for diagnosis and surgery. However, accurate polyp segmentation is a challenging task, for two major reasons: (i) the same type of polyps has a diversity of size, color and texture; and (ii) the boundary between a polyp and its surrounding mucosa is not sharp. To address these challenges, we propose a parallel reverse attention network (PraNet) for accurate polyp segmentation in colonoscopy images. Specifically, we first aggregate the features in high-level layers using a parallel partial decoder (PPD). Based on the combined feature, we then generate a global map as the initial guidance area for the following components. In addition, we mine the boundary cues using a reverse attention (RA) module, which is able to establish the relationship between areas and boundary cues. Thanks to the recurrent cooperation mechanism between areas and boundaries, our PraNet is capable of calibrating any misaligned predictions, improving the segmentation accuracy. Quantitative and qualitative evaluations on five challenging datasets across six metrics show that our PraNet improves the segmentation accuracy significantly, and presents a number of advantages in terms of generalizability, and real-time segmentation efficiency.

Topics

segmentation methodshigh resolution image

Keywords

Parallel Reverse Attention Networkpolyp segmentationbiomedical image segmentationparallel partial decodersalient object detectioncolonoscopy imagesdeep supervision

Table of Contents

  • 1 Introduction
  • 2 Method
  • 2.1 Feature Aggregating via Parallel Partial Decoder
  • 2.2 Reverse Attention Module
  • 2.3 Learning Process and Implementation Details.
  • 3 Experiments
  • 3.1 Experiments on Polyp Segmentation
  • 3.2 Ablation Study
  • 4 Conclusion
  • References

Knowls

  1. Knowl 1 — Parallel Reverse Attention Network (PraNet) Architecture

    model/method

    The Parallel Reverse Attention Network (PraNet) segments polyps in colonoscopy images by first establishing a coarse global semantic map and subsequently refining polyp boundaries via reverse attention.

    Given an input image II of size h×wh \times w, a Res2Net-based backbone extracts a 5-level feature hierarchy:

    {fi∣i=1,…,5}\{f_i \mid i = 1, \dots, 5\}

    where feature fif_i has spatial resolution [h/2i−1,w/2i−1][h / 2^{i-1}, w / 2^{i-1}]. Features are partitioned into low-level features {f1,f2}\{f_1, f_2\} and high-level features {f3,f4,f5}\{f_3, f_4, f_5\}. To conserve computational resources while leveraging high-level semantic context, low-level features are bypassed, and a Parallel Partial Decoder (pd(⋅)pd(\cdot)) aggregates high-level features via parallel connections:

    PD=pd(f3,f4,f5)PD = pd(f_3, f_4, f_5)

    This produces a high-level semantic global guidance map SgS_g. The feature representations and boundary cues are then progressively calibrated in a top-down manner through three cascading Reverse Attention (RA) modules operating on features f5f_5, f4f_4, and f3f_3, yielding side-output saliency maps S5S_5, S4S_4, and S3S_3. The final polyp segmentation prediction mask SpS_p is generated from the shallowest high-level side-output S3S_3 through a sigmoid operation: Sp=σ(S3)S_p = \sigma(S_3).

  2. Knowl 2 — Reverse Attention Module Formulation

    model/method

    The Reverse Attention (RA) module refines polyp boundaries by sequentially erasing currently estimated foreground polyp regions from high-level side-output features, forcing the network to discover missing details and refine misaligned boundary regions.

    For high-level side-output feature maps fi∈RCi×Hi×Wif_i \in \mathbb{R}^{C_i \times H_i \times W_i} where i∈{3,4,5}i \in \{3, 4, 5\}, the refined reverse attention feature RiR_i is computed via element-wise multiplication (⊙\odot) with a reverse attention weight matrix AiA_i:

    Ri=fi⊙AiR_i = f_i \odot A_i

    The reverse attention weight AiA_i is derived from the deeper prediction map Si+1S_{i+1} (with S6=SgS_6 = S_g) as:

    Ai=⊖(σ(P(Si+1)))A_i = \ominus\left(\sigma\left(P(S_{i+1})\right)\right)

    where P(⋅)P(\cdot) represents an up-sampling operation to match the spatial resolution of fif_i, σ(⋅)\sigma(\cdot) denotes the element-wise sigmoid activation function mapping values to [0,1][0, 1], and ⊖(⋅)\ominus(\cdot) denotes the reverse operation subtracting each element from a matrix of ones E∈R1×Hi×WiE \in \mathbb{R}^{1 \times H_i \times W_i} (Ai=E−σ(P(Si+1))A_i = E - \sigma(P(S_{i+1}))). The feature RiR_i is then processed via convolutional layers and residual connections to produce the updated prediction map SiS_i.

  3. Knowl 3 — PraNet Deep Supervision Multi-Scale Loss

    equation

    PraNet is trained end-to-end using deep supervision applied across the global guidance map SgS_g and the three high-level side-outputs {S3,S4,S5}\{S_3, S_4, S_5\}. The total training loss LtotalL_{total} is defined as:

    Ltotal=L(G,Sgup)+∑i=35L(G,Siup)L_{total} = L(G, S_g^{up}) + \sum_{i=3}^{5} L(G, S_i^{up})

    where GG is the binary ground-truth segmentation mask, and SgupS_g^{up} and SiupS_i^{up} denote the prediction maps SgS_g and SiS_i bilinearly upsampled to the spatial dimensions of GG.

    The compound loss function L(G,S)L(G, S) combines a weighted Intersection-over-Union loss LIoUwL_{IoU}^w and a weighted Binary Cross-Entropy loss LBCEwL_{BCE}^w:

    L(G,S)=LIoUw(G,S)+LBCEw(G,S)L(G, S) = L_{IoU}^w(G, S) + L_{BCE}^w(G, S)

    LIoUwL_{IoU}^w provides global shape restriction while assigning greater weight to hard pixels to emphasize challenging boundary regions. LBCEwL_{BCE}^w acts as a local pixel-level restriction, dynamically weighting hard pixels over easily classified background/foreground pixels.

  4. Knowl 4 — Implementation and Training Configuration for PraNet

    experimental setup

    PraNet is implemented in PyTorch using a Res2Net backbone and trained on an NVIDIA TITAN RTX GPU (24GB memory) with an Intel i9-9820X CPU.

    Input colonoscopy images are uniformly resized to 352×352352 \times 352 pixels. Training utilizes multi-scale inputs with scaling ratios {0.75,1.0,1.25}\{0.75, 1.0, 1.25\} without standard spatial/color data augmentation. Optimization is performed using the Adam optimizer with a fixed learning rate of 1×10−41 \times 10^{-4} and a batch size of 16. The full network is trained end-to-end for 20 epochs, reaching convergence in approximately 30 to 32 minutes.

    For standard benchmark evaluation, the Kvasir and CVC-ClinicDB (CVC-612) datasets are randomly partitioned into 80% for training, 10% for validation, and 10% for testing. Models trained on this combined split are evaluated directly on the test splits of seen datasets (Kvasir, CVC-612) as well as zero-shot across unseen datasets (CVC-ColonDB, ETIS, EndoScene-CVC300 test set).

  5. Knowl 5 — Quantitative Segmentation Performance on Seen Datasets

    data/table

    On the test splits of seen training datasets (Kvasir and CVC-612/CVC-ClinicDB), PraNet was compared against state-of-the-art medical segmentation baselines (U-Net, U-Net++, ResUNet-mod, ResUNet++, SFA). Six evaluation metrics were used: mean Dice, mean Intersection-over-Union (mean IoU), weighted Dice (FβwF_\beta^w), Structure-measure (SαS_\alpha), Enhanced-alignment metric (EϕmaxE_\phi^{max}), and Mean Absolute Error (MAE).

    Methods mean Dice mean IoU FβwF_\beta^w SαS_\alpha EϕmaxE_\phi^{max} MAE
    Kvasir
    U-Net 0.818 0.746 0.794 0.858 0.893 0.055
    U-Net++ 0.821 0.743 0.808 0.862 0.910 0.048
    ResUNet-mod 0.791 n/a n/a n/a n/a n/a
    ResUNet++ 0.813 0.793 n/a n/a n/a n/a
    SFA 0.723 0.611 0.670 0.782 0.849 0.075
    PraNet (Ours) 0.898 0.840 0.885 0.915 0.948 0.030
    CVC-612
    U-Net 0.823 0.755 0.811 0.889 0.954 0.019
    U-Net++ 0.794 0.729 0.785 0.873 0.931 0.022
    ResUNet-mod 0.779 n/a n/a n/a n/a n/a
    ResUNet++ 0.796 0.796 n/a n/a n/a n/a
    SFA 0.700 0.607 0.647 0.793 0.885 0.042
    PraNet (Ours) 0.899 0.849 0.896 0.936 0.979 0.009

    PraNet improved mean Dice by more than 7% absolute over the best competing method on both Kvasir (0.898 vs. 0.821) and CVC-612 (0.899 vs. 0.823) without any pre- or post-processing.

  6. Knowl 6 — Generalization Performance on Unseen Polyp Datasets

    data/table

    To evaluate generalizability, models trained on Kvasir and CVC-612 were tested without fine-tuning on three completely unseen datasets: CVC-ColonDB (380 images), ETIS (196 images), and the test set of EndoScene (CVC-T / CVC300).

    Methods mean Dice mean IoU FβwF_\beta^w SαS_\alpha EϕmaxE_\phi^{max} MAE
    CVC-ColonDB
    U-Net 0.512 0.444 0.498 0.712 0.776 0.061
    U-Net++ 0.483 0.410 0.467 0.691 0.760 0.064
    SFA 0.469 0.347 0.379 0.634 0.765 0.094
    PraNet (Ours) 0.709 0.640 0.696 0.819 0.869 0.045
    ETIS
    U-Net 0.398 0.335 0.366 0.684 0.740 0.036
    U-Net++ 0.401 0.344 0.390 0.683 0.776 0.035
    SFA 0.297 0.217 0.231 0.557 0.633 0.109
    PraNet (Ours) 0.628 0.567 0.600 0.794 0.841 0.031
    CVC-T (EndoScene)
    U-Net 0.710 0.627 0.684 0.843 0.876 0.022
    U-Net++ 0.707 0.624 0.687 0.839 0.898 0.018
    SFA 0.467 0.329 0.341 0.640 0.817 0.065
    PraNet (Ours) 0.871 0.797 0.843 0.925 0.972 0.010

    While competing methods (such as SFA and U-Net) suffered steep performance drops on out-of-distribution data, PraNet maintained superior generalization, exceeding the next-best model in mean Dice by 19.7% on CVC-ColonDB (0.709 vs. 0.512), 22.7% on ETIS (0.628 vs. 0.401), and 16.1% on CVC-T (0.871 vs. 0.710).

  7. Knowl 7 — Training Convergence and Real-Time Inference Speed Analysis

    data/table

    Training and inference performance were evaluated on the CVC-ClinicDB (CVC-612) benchmark under an identical hardware platform (Intel i9-9820X CPU and NVIDIA TITAN RTX GPU with 24GB memory).

    Methods Epoch Lr Training Inference mean Dice
    U-Net 30 3×10−43 \times 10^{-4} ∼\sim40 minutes ∼\sim8 fps 0.823
    U-Net++ 30 3×10−43 \times 10^{-4} ∼\sim45 minutes ∼\sim7 fps 0.794
    SFA 500 1×10−21 \times 10^{-2} >>20 hours ∼\sim40 fps 0.700
    PraNet (Ours) 20 1×10−41 \times 10^{-4} ∼\sim30 minutes ∼\sim50 fps 0.899

    PraNet achieved convergence in 20 epochs (approximately 30 minutes), significantly faster than SFA (>20 hours over 500 epochs), due to parallel short connections that efficiently back-propagate gradients to earlier decoder layers. During testing with 352×352352 \times 352 inputs, PraNet operated at ∼\sim50 fps, satisfying real-time colonoscopy video requirements.

  8. Knowl 8 — Ablation Study of Parallel Partial Decoder and Reverse Attention

    data/table

    The individual and combined contributions of the Parallel Partial Decoder (PPD) and Reverse Attention (RA) modules were evaluated using a Res2Net backbone on both a seen dataset (CVC-612) and an unseen dataset (CVC300) across mean Dice, mean IoU, and Structure-measure (SαS_\alpha).

    Settings CVC-612 (seen) CVC300 (unseen)
    mean Dice mean IoU SαS_\alpha mean Dice mean IoU SαS_\alpha
    No.1: Backbone 0.747 0.668 0.735 0.726 0.631 0.670
    No.2: PPD + Backbone 0.865 0.798 0.902 0.824 0.734 0.893
    No.3: RA + Backbone 0.888 0.845 0.912 0.871 0.800 0.888
    No.4: PPD + RA + Backbone 0.899 0.849 0.936 0.871 0.797 0.925

    Adding PPD to the backbone (No.2) improved mean Dice from 0.747 to 0.865 on CVC-612. Adding RA (No.3) boosted mean Dice to 0.888 on CVC-612 and from 0.726 to 0.871 on CVC300. Combining both in PraNet (No.4) achieved the overall best performance across all metrics on both seen and unseen datasets.

Coverage note — No substantial contributed material was omitted. All architectural modules (PPD, RA), loss formulations (weighted IoU and weighted BCE with deep supervision), experimental settings, comparative evaluations across all 5 benchmark datasets, inference efficiency benchmarks, and ablation studies are fully covered.

References

  1. 1.Akbari, M., Mohrekesh, M., Nasr-Esfahani, E., Soroushmehr, S.R., Karimi, N., Samavi, S., Najarian, K.: Polyp segmentation in colonoscopy images using fully convolutional network. In: IEEE EMBC. pp. 69–72 (2018)
  2. 2.Bernal, J., Sánchez, F.J., Fernández-Esparrach, G., Gil, D., Rodríguez, C., Vilariño, F.: Wm-dova maps for accurate polyp highlighting in colonoscopy: Validation vs. saliency maps from physicians. CMIG 43, 99–111 (2015)
  3. 3.Brandao, P., Mazomenos, E., Ciuti, G., Caliò, R., Bianchi, F., Menciassi, A., Dario, P., Koulaouzidis, A., Arezzo, A., Stoyanov, D.: Fully convolutional neural networks for polyp segmentation in colonoscopy. In: Medical Imaging 2017: Computer-Aided Diagnosis. vol. 10134, p. 101340F (2017)
  4. 4.Chen, S., Tan, X., Wang, B., Hu, X.: Reverse attention for salient object detection. In: ECCV. pp. 234–250 (2018)
  5. 5.Fan, D.P., Cheng, M.M., Liu, Y., Li, T., Borji, A.: Structure-measure: A new way to evaluate foreground maps. In: IEEE ICCV. pp. 4548–4557 (2017)
  6. 6.Fan, D.P., Gong, C., Cao, Y., Ren, B., Cheng, M.M., Borji, A.: Enhanced-alignment Measure for Binary Foreground Map Evaluation. In: IJCAI (2018)
  7. 7.Fan, D.P., Ji, G.P., Sun, G., Cheng, M.M., Shen, J., Shao, L.: Camouflaged object detection. In: IEEE CVPR (2020)
  8. 8.Fan, D.P., Liu, J.J., Gao, S.H., Hou, Q., Borji, A., Cheng, M.M.: Salient objects in clutter: Bringing salient object detection to the foreground. In: ECCV. pp. 1597–1604. Springer (2018)
  9. 9.Fan, D.P., Zhou, T., Ji, G.P., Zhou, Y., Chen, G., Fu, H., Shen, J., Shao, L.: Inf-Net: Automatic COVID-19 Lung Infection Segmentation from CT Images. IEEE TMI (2020)
  10. 10.Fang, Y., Chen, C., Yuan, Y., Tong, K.y.: Selective feature aggregation network with area-boundary constraints for polyp segmentation. In: MICCAI. pp. 302–310. Springer (2019)
  11. 11.Fu, K., Fan, D.P., Ji, G.P., Zhao, Q.: Jl-dcf: Joint learning and densely-cooperative fusion framework for rgb-d salient object detection. In: IEEE CVPR. pp. 3052–3062 (2020)
  12. 12.Gao, S.H., Cheng, M.M., Zhao, K., Zhang, X.Y., Yang, M.H., Torr, P.: Res2net: A new multi-scale backbone architecture. IEEE TPAMI pp. 1–1 (2020)
  13. 13.Gu, Z., Cheng, J., Fu, H., Zhou, K., Hao, H., Zhao, Y., Zhang, T., Gao, S., Liu, J.: CE-Net: Context encoder network for 2d medical image segmentation. IEEE TMI 38(10), 2281–2292 (2019)
  14. 14.Haggar, F.A., Boushey, R.P.: Colorectal cancer epidemiology: incidence, mortality, survival, and risk factors. Clinics in colon and rectal surgery 22(04), 191–197 (2009)
  15. 15.Jha, D., Smedsrud, P.H., Riegler, M.A., Halvorsen, P., de Lange, T., Johansen, D., Johansen, H.D.: Kvasir-seg: A segmented polyp dataset. In: MMM. pp. 451–462. Springer (2020)
  16. 16.Jha, D., Smedsrud, P.H., Riegler, M.A., Johansen, D., De Lange, T., Halvorsen, P., Johansen, H.D.: Resunet++: An advanced architecture for medical image segmentation. In: IEEE ISM. pp. 225–2255 (2019)
  17. 17.Jia, X., Xing, X., Yuan, Y., Xing, L., Meng, M.Q.H.: Wireless capsule endoscopy: A new tool for cancer screening in the colon with deep-learning-based polyp recognition. Proceedings of the IEEE 108(1), 178–197 (2019)
  18. 18.Mamonov, A.V., Figueiredo, I.N., Figueiredo, P.N., Tsai, Y.H.R.: Automated polyp detection in colon capsule endoscopy. IEEE TMI 33(7), 1488–1502 (2014)
  19. 19.Murugesan, B., Sarveswaran, K., Shankaranarayana, S.M., Ram, K., Joseph, J., Sivaprakasam, M.: Psi-Net: Shape and boundary aware joint multi-task deep network for medical image segmentation. In: IEEE EMBC. pp. 7223–7226 (2019)
  20. 20.Pogorelov, K., Randel, K.R., Griwodz, C., Eskeland, S.L., de Lange, T., Johansen, D., Spampinato, C., Dang-Nguyen, D.T., Lux, M., Schmidt, P.T., et al.: Kvasir: A multi-class image dataset for computer aided gastrointestinal disease detection. In: ACM MSC. pp. 164–169 (2017)
  21. 21.Qin, X., Zhang, Z., Huang, C., Gao, C., Dehghan, M., Jagersand, M.: Basnet: Boundary-aware salient object detection. In: IEEE CVPR. pp. 7479–7489 (2019)
  22. 22.Ronneberger, O., Fischer, P., Brox, T.: U-Net: Convolutional networks for biomedical image segmentation. In: MICCAI. pp. 234–241. Springer (2015)
  23. 23.Silva, J., Histace, A., Romain, O., Dray, X., Granado, B.: Toward embedded detection of polyps in wce images for early diagnosis of colorectal cancer. International Journal of Computer Assisted Radiology and Surgery 9(2), 283–293 (2014)
  24. 24.Tajbakhsh, N., Gurudu, S.R., Liang, J.: Automated polyp detection in colonoscopy videos using shape and context information. IEEE TMI 35(2), 630–644 (2015)
  25. 25.Vázquez, D., Bernal, J., Sánchez, F.J., Fernández-Esparrach, G., López, A.M., Romero, A., Drozdzal, M., Courville, A.: A benchmark for endoluminal scene segmentation of colonoscopy images. Journal of Healthcare Engineering 2017 (2017)
  26. 26.Wei, J., Wang, S., Huang, Q.: F3Net: Fusion, Feedback and Focus for Salient Object Detection. In: AAAI (2020)
  27. 27.Wei, Y., Feng, J., Liang, X., Cheng, M.M., Zhao, Y., Yan, S.: Object region mining with adversarial erasing: A simple classification to semantic segmentation approach. In: IEEE CVPR. pp. 1568–1576 (2017)
  28. 28.Wu, Y.H., Gao, S.H., Mei, J., Xu, J., Fan, D.P., Zhao, C.W., Cheng, M.M.: JCS: An Explainable COVID-19 Diagnosis System by Joint Classification and Segmentation. arXiv preprint arXiv:2004.07054 (2020)
  29. 29.Wu, Z., Su, L., Huang, Q.: Cascaded partial decoder for fast and accurate salient object detection. In: IEEE CVPR. pp. 3907–3916 (2019)
  30. 30.Yu, L., Chen, H., Dou, Q., Qin, J., Heng, P.A.: Integrating online and offline three-dimensional deep learning for automated polyp detection in colonoscopy videos. IEEE JBHI 21(1), 65–75 (2016)
  31. 31.Zhang, J., Fan, D.P., Dai, Y., Anwar, S., Sadat Saleh, F., Zhang, T., Barnes, N.: UC-Net: Uncertainty Inspired RGB-D Saliency Detection via Conditional Variational Autoencoders. In: IEEE CVPR (2020)
  32. 32.Zhang, R., Zheng, Y., Poon, C.C., Shen, D., Lau, J.Y.: Polyp detection during colonoscopy using a regression-based convolutional neural network with a tracker. Pattern Recognition 83, 209–219 (2018)
  33. 33.Zhang, S., Fu, H., Yan, Y., Zhang, Y., Wu, Q., Yang, M., Tan, M., Xu, Y.: Attention Guided Network for Retinal Image Segmentation. In: MICCAI, pp. 797–805 (2019)
  34. 34.Zhang, Z., Lin, Z., Xu, J., Jin, W., Lu, S.P., Fan, D.P.: Bilateral attention network for rgb-d salient object detection. arXiv preprint arXiv:2004.14582 (2020)
  35. 35.Zhang, Z., Liu, Q., Wang, Y.: Road extraction by deep residual u-net. IEEE Geoscience and Remote Sensing Letters 15(5), 749–753 (2018)
  36. 36.Zhang, Z., Fu, H., Dai, H., Shen, J., Pang, Y., Shao, L.: ET-Net: A generic edge-attention guidance network for medical image segmentation. In: MICCAI. pp. 442–450. Springer (2019)
  37. 37.Zhao, J.X., Cao, Y., Fan, D.P., Cheng, M.M., Li, X.Y., Zhang, L.: Contrast prior and fluid pyramid integration for rgbd salient object detection. In: IEEE CVPR. pp. 3927–3936 (2019)
  38. 38.Zhao, J.X., Liu, J.J., Fan, D.P., Cao, Y., Yang, J., Cheng, M.M.: EGNet: Edge guidance network for salient object detection. In: IEEE ICCV. pp. 8779–8788 (2019)
  39. 39.Zhou, Z., Siddiquee, M.M.R., Tajbakhsh, N., Liang, J.: Unet++: A nested u-net architecture for medical image segmentation. IEEE TMI pp. 3–11 (2019)

Citation

MLA
Fan, D.-P., et al. “PraNet: Parallel Reverse Attention Network for Polyp Segmentation”. arXiv, 2020, http://arxiv.org/abs/2006.11392v4.
APA
Fan, D.-P., Ji, G.-P., Zhou, T., Chen, G., Fu, H., Shen, J., & Shao, L. (2020). PraNet: Parallel Reverse Attention Network for Polyp Segmentation. arXiv. http://arxiv.org/abs/2006.11392v4
Chicago
Fan, D.-P., G.-P. Ji, T. Zhou, et al. 2020. “PraNet: Parallel Reverse Attention Network for Polyp Segmentation”. arXiv. http://arxiv.org/abs/2006.11392v4.
Harvard
Fan, D.-P. et al. (2020) “PraNet: Parallel Reverse Attention Network for Polyp Segmentation”, arXiv [Preprint]. Available at: http://arxiv.org/abs/2006.11392v4.
Vancouver
1. Fan D-P, Ji G-P, Zhou T, Chen G, Fu H, Shen J, Shao L (2020) PraNet: Parallel Reverse Attention Network for Polyp Segmentation. arXiv

BibTeX

@article{fan2020pranet,
  title = {PraNet: Parallel Reverse Attention Network for Polyp Segmentation},
  author = {Fan, Deng-Ping and Ji, Ge-Peng and Zhou, Tao and Chen, Geng and Fu, Huazhu and Shen, Jianbing and Shao, Ling},
  year = {2020},
  journal = {arXiv},
  url = {http://arxiv.org/abs/2006.11392v4},
  eprint = {2006.11392}
}
Metadata:arXiv

Source Code

This paper has an official code repository available. Click below to access the source code.

View Repository

Access the Paper

This paper is available from its original source. Click below to access the PDF.

Open PDF