pith. sign in

arxiv: 2402.02286 · v4 · submitted 2024-02-03 · 💻 cs.CV · cs.AI· cs.LG

Attention-Mamba: A Mamba-Enhanced Multi-Scale Parallel Inference Network for Medical Image Segmentation

Yanhua Zhang , Ke Zhang , Jingyu Wang , Gabriella Balestra , Samanta Rosati , Yulin Wu , Wuwei Wang , Valentina Giannini This is my paper

Pith reviewed 2026-05-24 03:46 UTC · model grok-4.3

classification 💻 cs.CV cs.AIcs.LG
keywords medical image segmentationMambamulti-scale featuresstate space modelattention mechanismparallel networkU-shaped architecture
0
0 comments X

The pith

A parallel multi-scale network using Mamba for long-range dependencies outperforms U-Nets and Transformers on medical segmentation tasks with only 14 million parameters.

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

The paper establishes that a departure from U-shaped encoders to parallel branches at multiple scales, each enhanced by Mamba state-space modeling, can capture both fine spatial details and global context more effectively than prior CNN or Transformer designs. Key additions include dual-path lateral connections, a Recursive Alignment Module to restore details in low-resolution features, and a Mamba-driven attention block that fuses predictions across scales. If the claim holds, clinicians would obtain more accurate organ and lesion outlines on MRI, CT, and dermoscopy images without paying the usual price in memory or speed. The reported results place the model first on Synapse, ACDC, ISIC-2018, and PH2 while ranking second-lowest in parameter count and moderate in FLOPs.

Core claim

The Attention-Mamba architecture replaces the classic U-Net with parallel inference branches at different resolutions, aligns low-resolution features via the Recursive Alignment Module, processes them through parallel Mamba branches for hierarchical global representations, and fuses the resulting multi-scale predictions with a Mamba-based attention mechanism that operates along both channel and spatial dimensions.

What carries the argument

Mamba-enhanced parallel multi-scale inference network whose dual-path lateral connections, Recursive Alignment Module, and Mamba-based attention fusion together produce adaptive multi-scale predictions.

If this is right

  • Medical segmentation models can achieve higher accuracy at lower parameter counts by using linear-complexity state-space models instead of quadratic attention.
  • Multi-scale predictions can be fused more effectively when the fusion step itself uses the same long-range modeling primitive as the encoder branches.
  • Lateral connections between high- and low-level paths inside each scale preserve spatial detail that would otherwise be lost in downsampling.

Where Pith is reading between the lines

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

  • The same parallel-branch layout could be tested on volumetric 3D data where memory constraints are even tighter.
  • Because Mamba scales linearly with sequence length, the architecture may remain practical at higher input resolutions than Transformer counterparts.
  • The Recursive Alignment Module's stepwise detail restoration might be reusable in other vision tasks that require precise localization after aggressive downsampling.

Load-bearing premise

The gains from the dual-path connections, alignment module, parallel Mamba branches, and Mamba attention arise from general improvements in feature handling rather than from tuning that works only on the four specific datasets tested.

What would settle it

A head-to-head comparison on a held-out medical imaging dataset from a new modality or acquisition protocol where the proposed model fails to exceed the best prior CNN or Transformer baseline in Dice score.

Figures

Figures reproduced from arXiv: 2402.02286 by Gabriella Balestra, Jingyu Wang, Ke Zhang, Samanta Rosati, Valentina Giannini, Wuwei Wang, Yanhua Zhang, Yulin Wu.

Figure 1
Figure 1. Figure 1: Inference speed and Params vs. mIoU accuracy on the Cityscapes [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The comparison between different parallel inference networks. The boxes represent feature maps, and their length and number roughly reflect the relative spatial resolution between them. (a) Independently passing the scaled images through the whole segmentation network (omitted for simplicity) to obtain multi-scale predictions. (b) The in-network parallel inference architecture proposed by Feature Pyramid N… view at source ↗
Figure 3
Figure 3. Figure 3: Overall architecture of our approach. (a) The details of MFAM. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The comparison between Straightforward Alignment and Recursive [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Long-range Skip Connection based Aggregation architecture (our [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Detailed design of Attention Head, Segmentation Head and Boundary [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative results on the valuation set of Cityscapes. Compared with [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Comparison of segmentation accuracy on di [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: The illustration of one of the ablations in Table. 8. All dashed boxes [PITH_FULL_IMAGE:figures/full_fig_p010_9.png] view at source ↗
read the original abstract

U-shaped architectures have long dominated the field of medical image segmentation, while Transformers are widely employed for modeling long-range dependencies. The former typically handles scale variations implicitly by aggregating multi-level features, whereas the efficiency of the latter is constrained by its quadratic computational and memory complexity. In this work, we propose an effective alternative to traditional U-shaped architectures by constructing parallel branches at different levels to obtain multi-scale features and corresponding predictions. Furthermore, we enhance our network by integrating Mamba, a state space model that captures long-range dependencies with linear complexity. First, a dual-path architecture with lateral connections aggregates high-level semantic information and low-level spatial details at each branch. Then, we introduce a Recursive Alignment Module (RAM) that restores spatial details in low-resolution features through stepwise alignment, optimizing them for subsequent global feature learning and multi-scale fusion. We further build parallel Mamba branches upon aligned features to establish hierarchical global representations. Finally, we propose a Mamba-based attention mechanism for adaptive multi-scale prediction fusion; this mechanism utilizes Mamba to enhance information exchange across scales along both the channel and spatial dimensions. Experiments across three imaging modalities (MRI, CT, and dermoscopy) underscore the superior generalization of the proposed network. Compared to state-of-the-art 2D CNN, Transformer, and Mamba-based networks, our model achieves the highest segmentation performance on the Synapse, ACDC, ISIC-2018, and PH2 datasets while maintaining high efficiency, featuring the second-smallest parameters (14.05 M) and moderate computational complexity (8.94 GFLOPs).

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 0 minor

Summary. The manuscript proposes Attention-Mamba, a multi-scale parallel inference network for medical image segmentation that replaces conventional U-shaped architectures with parallel branches at different scales. It integrates Mamba state-space models via a dual-path architecture with lateral connections, a Recursive Alignment Module (RAM) for restoring spatial details in low-resolution features, parallel Mamba branches for hierarchical global representations, and a Mamba-based attention mechanism for adaptive multi-scale fusion. The central empirical claim is that the model achieves state-of-the-art segmentation performance on the Synapse, ACDC, ISIC-2018, and PH2 datasets (across MRI, CT, and dermoscopy) while remaining efficient (second-smallest parameter count at 14.05 M and 8.94 GFLOPs).

Significance. If the reported performance gains and efficiency numbers prove robust under standard experimental protocols, the work would be significant for demonstrating a practical alternative to quadratic-complexity Transformers in medical segmentation. The combination of parallel multi-scale inference with linear-complexity long-range modeling via Mamba, plus the specific modules for alignment and fusion, could influence subsequent architectures that seek both accuracy and computational efficiency across imaging modalities.

major comments (2)
  1. [Abstract] Abstract: the claim that the proposed modules produce 'superior generalization' is not supported by the described experimental design. All quantitative results (including the headline SOTA numbers, 14.05 M parameters, and 8.94 GFLOPs) are confined to the four listed datasets (Synapse, ACDC, ISIC-2018, PH2) with no external validation cohort, cross-protocol transfer test, or multi-center evaluation reported. This leaves open the possibility that the gains from dual-path lateral connections, RAM, parallel Mamba branches, and Mamba-based attention fusion are dataset-specific rather than generalizable.
  2. [Abstract] Abstract: the central performance claim is presented without any reference to experimental protocol, statistical testing, ablation studies, or error bars. Because the abstract supplies only aggregate SOTA assertions and efficiency metrics, it is impossible to verify whether the reported superiority follows from the architectural innovations or from unstated differences in training procedure, data augmentation, or hyperparameter tuning.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major comment point by point below, proposing targeted revisions to the abstract where appropriate while maintaining the integrity of the reported experiments.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that the proposed modules produce 'superior generalization' is not supported by the described experimental design. All quantitative results (including the headline SOTA numbers, 14.05 M parameters, and 8.94 GFLOPs) are confined to the four listed datasets (Synapse, ACDC, ISIC-2018, PH2) with no external validation cohort, cross-protocol transfer test, or multi-center evaluation reported. This leaves open the possibility that the gains from dual-path lateral connections, RAM, parallel Mamba branches, and Mamba-based attention fusion are dataset-specific rather than generalizable.

    Authors: We agree that the phrasing 'superior generalization' in the abstract is not fully supported by the experimental scope, which is limited to the four datasets across three modalities without external cohorts or cross-protocol tests. The diversity of imaging modalities and datasets provides evidence of consistent performance, but we acknowledge this does not constitute broad generalization claims. We will revise the abstract to replace 'underscore the superior generalization' with 'demonstrate strong performance across multiple datasets and modalities' and add a brief limitations note in the discussion section. This constitutes a partial revision, as we cannot conduct new external validation experiments. revision: partial

  2. Referee: [Abstract] Abstract: the central performance claim is presented without any reference to experimental protocol, statistical testing, ablation studies, or error bars. Because the abstract supplies only aggregate SOTA assertions and efficiency metrics, it is impossible to verify whether the reported superiority follows from the architectural innovations or from unstated differences in training procedure, data augmentation, or hyperparameter tuning.

    Authors: The abstract is a high-level summary, with full experimental protocols, training details, ablation studies (Section 4.3), and comparative results provided in the main text. However, we recognize the referee's point that the abstract could better contextualize the claims. We will revise the abstract to include a short clause noting that results follow standard training protocols with ablations and comparisons detailed in the paper. This addresses the concern directly. revision: yes

Circularity Check

0 steps flagged

No circularity in architectural derivation

full rationale

The paper proposes a multi-scale parallel network with dual-path lateral connections, RAM, parallel Mamba branches, and Mamba-based attention fusion as explicit architectural constructions. No equations, fitted parameters, or self-citations are shown that would reduce any claimed performance gain or module output to a tautological redefinition of its inputs. The derivation chain consists of standard engineering choices justified by empirical results on four datasets rather than by self-referential definitions or imported uniqueness results. This is the normal case of a non-circular empirical architecture paper.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no explicit free parameters, axioms, or invented entities can be extracted beyond the implicit architectural decisions that define the new modules.

pith-pipeline@v0.9.0 · 5848 in / 1088 out tokens · 27100 ms · 2026-05-24T03:46:14.322663+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

88 extracted references · 88 canonical work pages · 8 internal anchors

  1. [1]

    Zhang, Y

    J. Zhang, Y . Shi, J. Yang, Q. Guo, Kd-scfnet: Towards more accurate and lightweight salient object detection via knowledge distillation, Neu- rocomputing 572 (2024) 127206

    work page 2024

  2. [2]

    S. Chen, P. Sun, Y . Song, P. Luo, Diffusiondet: Diffusion model for object detection, in: Proceedings of the IEEE /CVF International Conference on Computer Vision, 2023, pp. 19830–19843

    work page 2023

  3. [3]

    Zhang, L

    Y . Zhang, L. Feng, H. Shan, L. Yang, Z. Zhu, An aer-based spiking con- volution neural network system for image classification with low latency and high energy efficiency, Neurocomputing 564 (2024) 126984

    work page 2024

  4. [4]

    K. I. Rashid, C. Yang, C. Huang, Fast-dsagcn: Enhancing semantic segmentation with multifaceted attention mechanisms, Neurocomputing (2024) 127625

    work page 2024

  5. [5]

    J. Chen, J. Lu, X. Zhu, L. Zhang, Generative semantic segmentation, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pat- tern Recognition, 2023, pp. 7111–7120. 13

    work page 2023

  6. [6]

    de Jorge, R

    P. de Jorge, R. V olpi, P. H. Torr, G. Rogez, Reliability in semantic segmen- tation: Are we on the right track?, in: Proceedings of the IEEE/CVF Con- ference on Computer Vision and Pattern Recognition, 2023, pp. 7173– 7182

    work page 2023

  7. [7]

    Romera, J

    E. Romera, J. M. Alvarez, L. M. Bergasa, R. Arroyo, Erfnet: E fficient residual factorized convnet for real-time semantic segmentation, IEEE Trans. Intell. Transp. Syst. 19 (1) (2017) 263–272

    work page 2017

  8. [8]

    H. Zhao, X. Qi, X. Shen, J. Shi, J. Jia, Icnet for real-time semantic seg- mentation on high-resolution images, in: Proc. Eur. Conf. Comput. Vis. (ECCV), 2018, pp. 405–420

    work page 2018

  9. [9]

    L.-C. Chen, Y . Yang, J. Wang, W. Xu, A. L. Yuille, Attention to scale: Scale-aware semantic image segmentation, in: Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2016, pp. 3640–3649

    work page 2016

  10. [10]

    H. Zhao, J. Shi, X. Qi, X. Wang, J. Jia, Pyramid scene parsing network, in: Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2017, pp. 2881–2890

    work page 2017

  11. [11]

    Hariharan, P

    B. Hariharan, P. Arbel ´aez, R. Girshick, J. Malik, Hypercolumns for object segmentation and fine-grained localization, in: Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2015, pp. 447–456

    work page 2015

  12. [12]

    Orsic, I

    M. Orsic, I. Kreso, P. Bevandic, S. Segvic, In defense of pre-trained im- agenet architectures for real-time semantic segmentation of road-driving images, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit., 2019, pp. 12607–12616

    work page 2019

  13. [13]

    A. Tao, K. Sapra, B. Catanzaro, Hierarchical multi-scale attention for semantic segmentation, 2020, arXiv:2005.10821. [Online]. Available: https://arxiv.org/abs/2005.10821

    work page arXiv 2020

  14. [14]

    Zhang, K

    H. Zhang, K. Dana, J. Shi, Z. Zhang, X. Wang, A. Tyagi, A. Agrawal, Context encoding for semantic segmentation, in: Proc. IEEE Conf. Com- put. Vis. Pattern Recognit. (CVPR), 2018, pp. 7151–7160

    work page 2018

  15. [15]

    J. Wang, K. Sun, T. Cheng, B. Jiang, C. Deng, Y . Zhao, D. Liu, Y . Mu, M. Tan, X. Wang, et al., Deep high-resolution representation learning for visual recognition, IEEE Trans. Pattern Anal. Mach. Intell. 43 (10) (2020) 3349–3364

    work page 2020

  16. [16]

    T.-Y . Lin, P. Doll´ar, R. Girshick, K. He, B. Hariharan, S. Belongie, Fea- ture pyramid networks for object detection, in: Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2017, pp. 2117–2125

    work page 2017

  17. [17]

    D. Lin, Y . Ji, D. Lischinski, D. Cohen-Or, H. Huang, Multi-scale context intertwining for semantic segmentation, in: Proc. Eur. Conf. Comput. Vis. (ECCV), 2018, pp. 603–619

    work page 2018

  18. [18]

    Bilinski, V

    P. Bilinski, V . Prisacariu, Dense decoder shortcut connections for single- pass semantic segmentation, in: Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2018, pp. 6596–6605

    work page 2018

  19. [19]

    T. Xiao, Y . Liu, B. Zhou, Y . Jiang, J. Sun, Unified perceptual parsing for scene understanding, in: Proc. Eur. Conf. Comput. Vis. (ECCV), 2018, pp. 418–434

    work page 2018

  20. [20]

    Dvornik, K

    N. Dvornik, K. Shmelkov, J. Mairal, C. Schmid, Blitznet: A real-time deep network for scene understanding, in: Proc. IEEE Int. Conf. Comput. Vis. (ICCV), 2017, pp. 4154–4162

    work page 2017

  21. [21]

    H. Li, P. Xiong, H. Fan, J. Sun, Dfanet: Deep feature aggregation for real-time semantic segmentation, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit., 2019, pp. 9522–9531

    work page 2019

  22. [22]

    X. Weng, Y . Yan, G. Dong, C. Shu, B. Wang, H. Wang, J. Zhang, Deep multi-branch aggregation network for real-time semantic segmentation in street scenes, IEEE Trans. Intell. Transp. Syst. (2022)

    work page 2022

  23. [23]

    S. Liu, L. Qi, H. Qin, J. Shi, J. Jia, Path aggregation network for in- stance segmentation, in: Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2018, pp. 8759–8768

    work page 2018

  24. [24]

    Arani, S

    E. Arani, S. Marzban, A. Pata, B. Zonooz, Rgpnet: A real-time general purpose semantic segmentation, in: Proc. IEEE/CVF Winter Conf. Appl. Comput. Vis., 2021, pp. 3009–3018

    work page 2021

  25. [25]

    Kirillov, R

    A. Kirillov, R. Girshick, K. He, P. Doll ´ar, Panoptic feature pyramid net- works, in: Proc. IEEE/CVF Conf. Comput. Vis. Pattern Recognit., 2019, pp. 6399–6408

    work page 2019

  26. [26]

    W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.-Y . Fu, A. C. Berg, Ssd: Single shot multibox detector, in: Computer Vision–ECCV 2016: 14th European Conference, Amsterdam, The Netherlands, October 11–14, 2016, Proceedings, Part I 14, Springer, 2016, pp. 21–37

    work page 2016

  27. [27]

    B. Hu, J. Huang, Y . Liu, Y .-W. Tai, C.-K. Tang, Nerf-rpn: A general framework for object detection in nerfs, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2023, pp. 23528–23538

    work page 2023

  28. [28]

    R. Kaur, S. Singh, A comprehensive review of object detection with deep learning, Digital Signal Processing 132 (2023) 103812

    work page 2023

  29. [29]

    Z. Zou, K. Chen, Z. Shi, Y . Guo, J. Ye, Object detection in 20 years: A survey, Proceedings of the IEEE 111 (3) (2023) 257–276

    work page 2023

  30. [30]

    J. Long, E. Shelhamer, T. Darrell, Fully convolutional networks for se- mantic segmentation, in: Proc. IEEE Conf. Comput. Vis. Pattern Recog- nit. (CVPR), 2015, pp. 3431–3440

    work page 2015

  31. [31]

    Shuai, Z

    B. Shuai, Z. Zuo, B. Wang, G. Wang, Scene segmentation with dag- recurrent neural networks, IEEE Trans. Pattern Anal. Mach. Intell. 40 (6) (2017) 1480–1493

    work page 2017

  32. [32]

    C. Peng, X. Zhang, G. Yu, G. Luo, J. Sun, Large kernel matters–improve semantic segmentation by global convolutional network, in: Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2017, pp. 4353–4361

    work page 2017

  33. [33]

    H. Ding, X. Jiang, B. Shuai, A. Q. Liu, G. Wang, Semantic segmenta- tion with context encoding and multi-path decoding, IEEE Trans. Image Process. 29 (2020) 3520–3533

    work page 2020

  34. [34]

    X. Ding, C. Shen, T. Zeng, Y . Peng, Sab net: A semantic attention boost- ing framework for semantic segmentation, IEEE Trans. Neural Netw. Learn. Syst. (2022)

    work page 2022

  35. [35]

    Guided Upsampling Network for Real-Time Semantic Segmentation

    D. Mazzini, Guided upsampling network for real-time seman- tic segmentation, 2018, arXiv:1807.07466. [Online]. Available: http://arxiv.org/abs/1807.07466

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  36. [36]

    Mazzini, R

    D. Mazzini, R. Schettini, Spatial sampling network for fast scene un- derstanding, in: Proc. IEEE /CVF Conf. Comput. Vis. Pattern Recognit. Workshops (CVPRW), 2019, pp. 0–0

    work page 2019

  37. [37]

    Zhang, K

    Z. Zhang, K. Zhang, Farsee-net: Real-time semantic segmentation by ef- ficient multi-scale context aggregation and feature space super-resolution, in: Proc. IEEE Int. Conf. Rob Autom, IEEE, 2020, pp. 8411–8417

    work page 2020

  38. [38]

    L.-C. Chen, G. Papandreou, I. Kokkinos, K. Murphy, A. L. Yuille, Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs, IEEE Trans. Pattern Anal. Mach. Intell. 40 (4) (2017) 834–848

    work page 2017

  39. [39]

    L.-C. Chen, G. Papandreou, I. Kokkinos, K. Murphy, A.-L. Yuille, Se- mantic image segmentation with deep convolutional nets and fully con- nected crfs, arXiv preprint arXiv:1412.7062 (2014)

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  40. [40]

    Z. Yang, H. Yu, M. Feng, W. Sun, X. Lin, M. Sun, Z.-H. Mao, A. Mian, Small object augmentation of urban scenes for real-time semantic seg- mentation, IEEE Trans. Image Process. 29 (2020) 5175–5190

    work page 2020

  41. [41]

    Zhuang, J

    J. Zhuang, J. Yang, L. Gu, N. Dvornek, Shelfnet for fast semantic segmen- tation, in: Proc. IEEE/CVF Int. Conf. Comput. Vis. Workshop (ICCVW), 2019, pp. 0–0

    work page 2019

  42. [42]

    X. Li, A. You, Z. Zhu, H. Zhao, M. Yang, K. Yang, S. Tan, Y . Tong, Semantic flow for fast and accurate scene parsing, in: Proc. Eur. Conf. Comput. Vis. (ECCV), Springer, 2020, pp. 775–793

    work page 2020

  43. [43]

    S. Hao, Y . Zhou, Y . Guo, R. Hong, J. Cheng, M. Wang, Real-time se- mantic segmentation via spatial-detail guided context propagation, IEEE Trans. Neural Netw. Learn. Syst. (2022)

    work page 2022

  44. [44]

    D. Lin, D. Shen, S. Shen, Y . Ji, D. Lischinski, D. Cohen-Or, H. Huang, Zigzagnet: Fusing top-down and bottom-up context for object segmen- tation, in: Proc. IEEE /CVF Conf. Comput. Vis. Pattern Recognit., 2019, pp. 7490–7499

    work page 2019

  45. [45]

    H. Noh, S. Hong, B. Han, Learning deconvolution network for semantic segmentation, in: Proc. IEEE Int. Conf. Comput. Vis. (ICCV), 2015, pp. 1520–1528

    work page 2015

  46. [46]

    Badrinarayanan, A

    V . Badrinarayanan, A. Kendall, R. Cipolla, Segnet: A deep convolutional encoder-decoder architecture for image segmentation, IEEE Trans. Pat- tern Anal. Mach. Intell. 39 (12) (2017) 2481–2495

    work page 2017

  47. [47]

    Huang, Y

    Z. Huang, Y . Wei, X. Wang, W. Liu, T. S. Huang, H. Shi, Alignseg: Feature-aligned segmentation networks, IEEE Trans. Pattern Anal. Mach. Intell. 44 (1) (2021) 550–557

    work page 2021

  48. [48]

    Huang, Z

    S. Huang, Z. Lu, R. Cheng, C. He, Fapn: Feature-aligned pyramid net- work for dense image prediction, in: Proc. IEEE/CVF Int. Conf. Comput. Vis., 2021, pp. 864–873

    work page 2021

  49. [49]

    J. Dai, H. Qi, Y . Xiong, Y . Li, G. Zhang, H. Hu, Y . Wei, Deformable convolutional networks, in: Proc. IEEE Int. Conf. Comput. Vis. (ICCV), 2017, pp. 764–773

    work page 2017

  50. [50]

    Takikawa, D

    T. Takikawa, D. Acuna, V . Jampani, S. Fidler, Gated-scnn: Gated shape cnns for semantic segmentation, in: Proc. IEEE /CVF Int. Conf. Comput. Vis., 2019, pp. 5229–5238

    work page 2019

  51. [51]

    Borse, Y

    S. Borse, Y . Wang, Y . Zhang, F. Porikli, Inverseform: A loss function for structured boundary-aware segmentation, in: Proc. IEEE /CVF Conf. 14 Comput. Vis. Pattern Recognit., 2021, pp. 5901–5911

    work page 2021

  52. [52]

    Canny, A computational approach to edge detection, IEEE Trans

    J. Canny, A computational approach to edge detection, IEEE Trans. Pat- tern Anal. Mach. Intell. (6) (1986) 679–698

    work page 1986

  53. [53]

    Shrivastava, A

    A. Shrivastava, A. Gupta, R. Girshick, Training region-based object de- tectors with online hard example mining, in: Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2016, pp. 761–769

    work page 2016

  54. [54]

    C. Yu, J. Wang, C. Peng, C. Gao, G. Yu, N. Sang, Bisenet: Bilateral segmentation network for real-time semantic segmentation, in: Proc. Eur. Conf. Comput. Vis. (ECCV), 2018, pp. 325–341

    work page 2018

  55. [55]

    Cordts, M

    M. Cordts, M. Omran, S. Ramos, T. Rehfeld, M. Enzweiler, R. Benen- son, U. Franke, S. Roth, B. Schiele, The cityscapes dataset for semantic urban scene understanding, in: Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2016, pp. 3213–3223

    work page 2016

  56. [56]

    G. J. Brostow, J. Fauqueur, R. Cipolla, Semantic object classes in video: A high-definition ground truth database, Pattern Recognit. Lett. 30 (2) (2009) 88–97

    work page 2009

  57. [57]

    Sturgess, K

    P. Sturgess, K. Alahari, L. Ladicky, P. H. Torr, Combining appearance and structure from motion features for road scene understanding, in: Proc. BMVC, 2009

    work page 2009

  58. [58]

    Mottaghi, X

    R. Mottaghi, X. Chen, X. Liu, N.-G. Cho, S.-W. Lee, S. Fidler, R. Urta- sun, A. Yuille, The role of context for object detection and semantic seg- mentation in the wild, in: Proc. IEEE Conf. Comput. Vis. Pattern Recog- nit. (CVPR), 2014, pp. 891–898

    work page 2014

  59. [59]

    J. Xie, B. Shuai, J.-F. Hu, J. Lin, W.-S. Zheng, Improving fast segmen- tation with teacher-student learning, 2018, arXiv:1810.08476. [Online]. Available: http://arxiv.org/abs/1810.08476

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  60. [60]

    H. Ding, X. Jiang, A. Q. Liu, N. M. Thalmann, G. Wang, Boundary- aware feature propagation for scene segmentation, in: Proc. IEEE /CVF Int. Conf. Comput. Vis., 2019, pp. 6819–6829

    work page 2019

  61. [61]

    T. He, C. Shen, Z. Tian, D. Gong, C. Sun, Y . Yan, Knowledge adaptation for efficient semantic segmentation, in: Proc. IEEE /CVF Conf. Comput. Vis. Pattern Recognit., 2019, pp. 578–587

    work page 2019

  62. [62]

    Or ˇsi´c, S

    M. Or ˇsi´c, S. ˇSegvi´c, Efficient semantic segmentation with pyramidal fu- sion, Pattern Recognit. 110 (2021) 107611

    work page 2021

  63. [63]

    ENet: A Deep Neural Network Architecture for Real-Time Semantic Segmentation

    A. Paszke, A. Chaurasia, S. Kim, E. Culurciello, Enet: A deep neu- ral network architecture for real-time semantic segmentation, 2016, arXiv:1606.02147. [Online]. Available: http://arxiv.org/abs/1606.02147

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  64. [64]

    Mehta, M

    S. Mehta, M. Rastegari, A. Caspi, L. Shapiro, H. Hajishirzi, Espnet: E ffi- cient spatial pyramid of dilated convolutions for semantic segmentation, in: Proc. Eur. Conf. Comput. Vis. (ECCV), 2018, pp. 552–568

    work page 2018

  65. [65]

    R. P. Poudel, U. Bonde, S. Liwicki, C. Zach, Contextnet: Explor- ing context and detail for semantic segmentation in real-time, 2018, arXiv:1805.04554. [Online]. Available: http://arxiv.org/abs/1805.04554

    work page internal anchor Pith review Pith/arXiv arXiv 2018

  66. [66]

    R. P. Poudel, S. Liwicki, R. Cipolla, Fast-scnn: Fast semantic segmentation network, 2019, arXiv:1902.04502. [Online]. Available: http://arxiv.org/abs/1902.04502

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  67. [67]

    Lo, H.-M

    S.-Y . Lo, H.-M. Hang, S.-W. Chan, J.-J. Lin, E fficient dense modules of asymmetric convolution for real-time semantic segmentation, in: Int. Conf. Multimed. Asia, 2019, pp. 1–6

    work page 2019

  68. [68]

    T. Wu, S. Tang, R. Zhang, J. Cao, Y . Zhang, Cgnet: A light-weight context guided network for semantic segmentation, IEEE Trans. Image Process. 30 (2020) 1169–1179

    work page 2020

  69. [69]

    Jiang, W

    B. Jiang, W. Tu, C. Yang, J. Yuan, Context-integrated and feature-refined network for lightweight object parsing, IEEE Trans. Image Process. 29 (2020) 5079–5093

    work page 2020

  70. [70]

    Alonso, L

    I. Alonso, L. Riazuelo, A. C. Murillo, Mininet: An e fficient semantic seg- mentation convnet for real-time robotic applications, IEEE Trans. Robot. 36 (4) (2020) 1340–1347

    work page 2020

  71. [71]

    Q. Zhou, Y . Wang, Y . Fan, X. Wu, S. Zhang, B. Kang, L. J. Latecki, Aglnet: Towards real-time semantic segmentation of self-driving images via attention-guided lightweight network, Appl. Soft. Comput. 96 (2020) 106682

    work page 2020

  72. [72]

    C. Yu, C. Gao, J. Wang, G. Yu, C. Shen, N. Sang, Bisenet v2: Bilat- eral network with guided aggregation for real-time semantic segmenta- tion, Int. J. Comput. Vis. 129 (11) (2021) 3051–3068

    work page 2021

  73. [73]

    M. Fan, S. Lai, J. Huang, X. Wei, Z. Chai, J. Luo, X. Wei, Rethinking bisenet for real-time semantic segmentation, in: Proc. IEEE /CVF Conf. Comput. Vis. Pattern Recognit., 2021, pp. 9716–9725

    work page 2021

  74. [74]

    G. Lin, A. Milan, C. Shen, I. Reid, Refinenet: Multi-path refinement net- works for high-resolution semantic segmentation, in: Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2017, pp. 1925–1934

    work page 2017

  75. [75]

    Zhang, S

    R. Zhang, S. Tang, Y . Zhang, J. Li, S. Yan, Scale-adaptive convolutions for scene parsing, in: Proc. IEEE Int. Conf. Comput. Vis. (ICCV), 2017, pp. 2031–2039

    work page 2017

  76. [76]

    S. Kong, C. C. Fowlkes, Recurrent scene parsing with perspective under- standing in the loop, in: Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2018, pp. 956–965

    work page 2018

  77. [77]

    Z. Wu, C. Shen, A. Van Den Hengel, Wider or deeper: Revisiting the resnet model for visual recognition, Pattern Recognit. 90 (2019) 119–133

    work page 2019

  78. [78]

    C. Yu, J. Wang, C. Peng, C. Gao, G. Yu, N. Sang, Learning a discrimi- native feature network for semantic segmentation, in: Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2018, pp. 1857–1866

    work page 2018

  79. [79]

    H. Zhao, Y . Zhang, S. Liu, J. Shi, C. C. Loy, D. Lin, J. Jia, Psanet: Point- wise spatial attention network for scene parsing, in: Proc. Eur. Conf. Comput. Vis. (ECCV), 2018, pp. 267–283

    work page 2018

  80. [80]

    J. Fu, J. Liu, H. Tian, Y . Li, Y . Bao, Z. Fang, H. Lu, Dual attention network for scene segmentation, in: Proc. IEEE /CVF Conf. Comput. Vis. Pattern Recognit., 2019, pp. 3146–3154

    work page 2019

Showing first 80 references.