Pith. sign in

REVIEW 6 major objections 6 minor 50 references

Threshold Attention Network for Semantic Segmentation of Remote Sensing Images

T0 review · 6 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read This paper claims that attention between groups of pixels sharing similar feature values can replace pixel-pair attention in semantic segmentation, cutting computational cost from quadratic to roughly linear while maintaining or improving…

desk verdict Threshold attention is a genuinely new mechanism, but the SOTA comparison is confounded by TTA/OHEM/aux loss and the efficiency claim needs qualification; worth review, not desk reject. read the letter →

arxiv 2501.07984 v1 pith:WTG3WNGT submitted 2025-01-14 cs.CV

Wei Long , Yongjun Zhang , Zhongwei Cui , Yujie Xu , Xuexue Zhang This is my paper · ORCID
classification cs.CV
keywords semanticsegmentationremotesensingimagerythresholdattentionmechanismself-attentionglobalcontextmodelingpyramidpoolingefficienthigh-resolutionurbanscenes
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to establish that a segmentation network can compute global attention between groups of pixels sharing similar feature values instead of between every pixel pair, and that this substitution both lowers cost and improves accuracy on high-resolution remote sensing images. The proposed threshold attention mechanism (TAM) quantizes each feature channel into $L$ value intervals, computes an $L\times L$ attention matrix between intervals, then reattaches each pixel to its interval through a one-hot location code. Wrapped into two modules—AFEM for shallow features and TAPP for deep multi-scale features—TAM forms TANet, which the paper reports achieves top or near-top mean F1, overall accuracy, and mIoU on the Vaihingen and Potsdam benchmarks while using roughly one-twelfth the GFLOPs of standard self-attention. The result matters because global context in segmentation has until now come with quadratic pixel-pair cost, limiting use on very large imagery.

What carries the argument

The load-bearing object is the threshold attention mechanism, an attention layer whose tokens are per-channel value intervals rather than pixels. For input features $F\in\mathbb{R}^{C\times N}$ with $N=H\times W$, the mechanism computes per-channel thresholds that split each channel into $L$ equal-width intervals, obtaining $T\in\mathbb{R}^{L\times C}$; it then applies query, key, and value projections and row-softmax to get an $L\times L$ attention matrix $A$, so the cost depends on $L$ rather than on $N^2$. Pixel locations are preserved by the discretization map $P\in\mathbb{R}^{C\times N}$, one-hot encoded and multiplied against the transposed attention matrix, which gives each pixel the attention weight of its bin. The same mechanism is reused in AFEM, paired with channel attention and a residual skip to enhance shallow features, and in TAPP, paired with dilated convolutions and global pooling to enrich deep multi-scale features.

What would settle it

Replace the threshold-level assignment in Eq. (7) with a random assignment of pixels to the same number $L$ of levels while keeping all other machinery fixed: if segmentation accuracy stays about the same, the value-based grouping itself is doing no work and the mechanism is just a cheap low-rank attention. The paper's own observation that the tree class is its weakest, attributed to thin branches and wide color variation, offers a second check: on classes with high intra-class color spread, TAM should systematically lose to pixel-pair attention, and a per-class accuracy comparison on those regions would expose the grouping failure.

Watch

Extended reading notes

Core claim

The central claim is that dependencies between spatially distant pixels can be modeled faithfully at the level of value bins: pixels whose feature vectors fall in the same per-channel interval belong to a 'threshold region,' and attention between these regions carries the global information the network needs. TAM performs this in three moves: uniform per-channel quantization to a threshold matrix $T\in\mathbb{R}^{L\times C}$, linear projections and softmax to produce an $L\times L$ attention matrix, and one-hot level assignment $P$ that scatters the region-level attention back onto individual pixels so the output has the same shape as the input. The paper reports that TANet built from AFEM and TAPP exceeds the listed comparison methods on most overall metrics, and that the two threshold-attention modules outweigh classical context and attention modules in the Vaihingen ablation.

Load-bearing premise

The method assumes that pixels falling in the same per-channel value interval are similar enough semantically that attention between intervals can replace attention between pixels; if binning destroys the correlations needed for accurate segmentation, the efficiency gain is bought at the cost of accuracy.

Editorial extensions

If this is right

  • Global attention for segmentation becomes affordable at high resolution: TAM's cost scales with the chosen number of value levels $L$ and linearly with pixel count, rather than quadratically.
  • The same modules can be dropped into other encoder-decoder segmentation networks because AFEM and TAPP preserve input and output shape and are reported to improve other standard backbones too.
  • The level count $L$ is a meaningful hyperparameter: optimal values differ between modules and datasets, so threshold granularity can be tuned per task.
  • Standard training add-ons such as auxiliary loss, online hard example mining, and test-time augmentation stack on top of TANet's gains, implying the region-level attention is complementary to loss- and inference-time improvements.

Reading between the lines

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

  • The paper does not test TAM against alternative grouping rules, such as clustering by learned prototypes or per-class thresholds; a natural follow-up is whether equal-width value bins are the best partition or merely a cheap one.
  • Because the efficiency comparison excludes the backbone and the reported per-image inference time is not the fastest, the practical speed advantage over other full networks may be smaller than the GFLOPs table suggests.
  • The mechanism's grouping premise suggests a stress test on images with strong illumination gradients or sensor noise, where equal-width bins may split one object across many levels; adaptive or normalized binning would be a testable fix.
  • The same quantization-then-attention pattern could apply to other dense prediction tasks such as panoptic segmentation or depth estimation, where global region dependencies matter; the paper only demonstrates semantic segmentation.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

6 major / 6 minor

Summary. The paper proposes a Threshold Attention Mechanism (TAM) that quantizes each channel of a feature map into L intervals, computes attention between the resulting L per-channel threshold features, and then reassigns the attention weights to pixels via a one-hot position matrix, achieving complexity that is linear in the number of pixels. Based on TAM, the authors build TANet, consisting of an Attentional Feature Enhancement Module (AFEM) for shallow features and a Threshold Attention Pyramid Pooling module (TAPP) for deep features. The method is evaluated on the ISPRS Vaihingen and Potsdam semantic segmentation benchmarks, reporting state-of-the-art or competitive mIoU, OA, and F1 scores.

Significance. If the claims are substantiated, the threshold attention idea is a potentially useful alternative to dense self-attention for high-resolution remote sensing segmentation, and the paper includes a reasonably extensive experimental study with parameter ablations and module-level efficiency comparisons. The manuscript also makes its core formulas explicit, which is helpful for reproducibility. However, the two headline claims — that TAM 'significantly reduces computational effort' and that TANet outperforms state-of-the-art methods — are currently undermined by internal inconsistencies and uncontrolled comparisons, as detailed below.

major comments (6)
  1. [§V-C, Tables X and XII] The state-of-the-art comparison is not controlled for training enhancements. The TANet entries in Tables XII and XIII use OHEM, auxiliary loss, and test-time augmentation; Table X shows these add +1.10 mIoU on Vaihingen (83.35→84.45). Without them, TANet's Vaihingen mIoU of 83.35 is below HMANet (83.49), MFNet (83.50), CTMFNet (84.34), and HBCNet (84.21) in the same table. The paper never states whether the cited baselines were trained or evaluated with these techniques. Please provide a controlled comparison, such as reporting TANet without OHEM/aux loss/TTA as the primary SOTA number, or reporting baselines with and without the same enhancements.
  2. [§V-B2, Tables VIII-IX and Abstract] The computational-complexity claim is internally inconsistent. Table VIII reports AFEM+TAPP at 49 GFLOPs versus 619 for standard self-attention, but Table IX shows TANet's end-to-end inference time (0.091 s) is slower than SA (0.076 s). The paper should clarify that the GFLOPs comparison is at the module level only, report end-to-end FLOPs and inference time for the full networks, and discuss why the module-level reduction does not translate into faster inference.
  3. [§III-B, Eq. (7)] The discretization formula uses a factor 2L, whereas Eq. (1) defines L threshold levels. As written, P can take 2L+1 distinct values (0,...,2L), but Eq. (8) requires P'_c ∈ R^{L×N} (L rows) for the one-hot encoding. This inconsistency breaks the TAM computation as described. Please correct Eq. (7) to use L (or adjust the one-hot dimension) and state the exact value range of P.
  4. [§IV-A1] The Vaihingen data split description is self-contradictory. The text says 16 training and 17 test regions, then lists 17 image IDs for training, and then says 'Thirty labeled images are used for validation,' which cannot be reconciled with a 33-image dataset. Please provide an exact, consistent split of training, validation, and test images.
  5. [§III-E] The OHEM parameter S is set to 10,000 but is never defined, and the description of OHEM via a probability threshold θ is nonstandard (OHEM typically selects a fixed number of hardest examples). Please define S and describe the hard-example selection procedure precisely.
  6. [§III-B] The core premise that per-channel intensity intervals correspond to semantically coherent pixel regions is not analyzed or compared with alternative grouping schemes such as adaptive binning, k-means, or superpixels. Because this premise underlies the claimed accuracy-efficiency advantage of TAM, please provide at least an ablation or analysis that isolates the effect of the quantization/grouping strategy.
minor comments (6)
  1. [Table X caption] The caption spells 'Vaihingen' as 'VAHINGEN'; please correct it.
  2. [§IV-A1] In the sentence 'the 15 images labeled as follows:', the list contains 17 IDs; please correct the count and the surrounding text.
  3. [§V-C] The term 'Test-Time Enhancement' should be 'Test-Time Augmentation' throughout the paper.
  4. [§IV-C] The AdamW optimizer's weight decay value is not reported; please state it for reproducibility.
  5. [§V-A] Please state explicitly whether the L1 and L2 parameters were selected on the validation set or on the test set, so readers can assess the parameter-study protocol.
  6. [Table XIII] In the CASIA3 row, the Building F1 score of 86.80 is anomalously low compared with the other rows; please verify it against the original source.

Circularity Check

1 steps flagged · score 2.0 of 10

No structural circularity in the TAM equations; the SOTA claim is weakened by selecting L1/L2 on the evaluation datasets, but the mechanism itself is an independent architecture claim.

  1. fitted input called prediction [Section IV-C (Implementation Details) and Section V-A (Parameter Study), Tables I-II and IV-V; final comparisons in Tables XII-XIII.]
    "The threshold number L of the threshold attention module was optimized for different datasets. ... We experimentally studied the effect of L on the segmentation performance of the network. ... It is evident that TANet obtains the optimal semantic segmentation performance when L1 is set to 150 and L2 is set to 200."

    The final SOTA tables report TANet using L1/L2 values chosen as the maxima of mIoU sweeps run on the same Vaihingen and Potsdam datasets that those tables evaluate. For example, Table I selects L1=150 by the top mIoU (83.35) on Vaihingen, and the final TANet row reproduces that same 83.35 before OHEM/Aux/TTA are added. The paper does not state that the sweeps used a held-out validation split, and the split description ('Thirty labeled images are used for validation') is inconsistent with the 33-image Vaihingen set. The reported advantage is therefore partly a selected maximum over the evaluation data rather than an independent prediction. This is an evaluation-protocol circularity; it does not make the TAM equations equivalent to their inputs.

full rationale

The paper's derivation chain is an architecture proposal, not a formal theorem. Equations (1)-(8) define TAM as per-channel min-max quantization into L levels, L-level dot-product attention, and reassignment of each pixel to its level's attention vector; the output is a learned reweighting of the input and is not equal to the input or to the fitted parameters by construction. The efficiency claim is supported by direct GFLOPS/parameter measurements (Table VIII), and the accuracy claim is an external benchmark comparison. No load-bearing self-citation is used: the only overlapping-author reference, HBCNet [24], serves as a comparison baseline. The one circularity-adjacent issue is that the threshold number L is explicitly 'optimized for different datasets' and the reported final results use the values that maximize the same reported metrics, with no clear held-out validation split. Because the central TAM design still has independent empirical content and the grid search is small, a score of 2 is appropriate rather than a higher fitted-input score.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central empirical claim rests on a small number of hyperparameters selected for each dataset and on an unproven modeling assumption that value quantization preserves the correlations needed for segmentation. No new physical entities are introduced; the TAM modules are architectural constructions built from existing operations.

free parameters (6)
  • L1 (AFEM threshold levels) = 150 on Vaihingen; 200 on Potsdam
    Grid-searched over six values per dataset, Tables I and IV; controls quantization granularity in AFEM.
  • L2 (TAPP threshold levels) = 200 on both datasets
    Grid-searched over six values per dataset, Tables II and V; controls quantization granularity in TAPP.
  • lambda (auxiliary loss weight) = 0.5
    Set based on limited experiments on Vaihingen as described in Section IV-C; balances auxiliary and main loss.
  • theta (OHEM threshold) = 0.65
    Set based on limited experiments on Vaihingen; used for hard-example selection but never formally defined in an equation.
  • S (unexplained parameter) = 10000
    Mentioned in Section IV-C as a selected parameter, but never defined or used in any equation.
  • Initial learning rate and poly power = 0.0005 and 0.9
    Standard poly schedule; initial value and power chosen by the authors without ablation.
assumptions (4)
  • domain assumption Self-attention computational complexity grows rapidly with pixel count, motivating a linear-complexity alternative.
    The paper states in the abstract and Section I that dense attention maps cause 'exponential' increases in complexity, when the actual scaling is quadratic; this premise drives the entire efficiency argument.
  • ad hoc to paper Pixels in the same per-channel threshold interval are semantically similar enough for grouped attention.
    Eqs. (1) and (7) define uniform threshold intervals; the paper assumes pixels in the same interval share semantic meaning, without proof or comparison to other grouping schemes.
  • ad hoc to paper Uniform quantization with L levels per channel is a sufficient representation of feature distribution.
    The choice of L equal-size intervals is assumed to be good; no alternative quantization such as k-means or learned bins is examined.
  • ad hoc to paper Hyperparameters tuned on Vaihingen transfer to other datasets.
    Lambda, theta, and S were selected via experiments on Vaihingen and applied to Potsdam without additional tuning; this transferability is assumed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Threshold Attention Network for Semantic Segmentation of Remote Sensing Images." pith.science (2026). https://pith.science/paper/WTG3WNGT

@misc{pith2026250107984,
  author       = {Pith},
  title        = {Pith review of: Threshold Attention Network for Semantic Segmentation of Remote Sensing Images},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WTG3WNGT}},
  note         = {Machine review of arXiv:2501.07984}
}
read the original abstract

Semantic segmentation of remote sensing images is essential for various applications, including vegetation monitoring, disaster management, and urban planning. Previous studies have demonstrated that the self-attention mechanism (SA) is an effective approach for designing segmentation networks that can capture long-range pixel dependencies. SA enables the network to model the global dependencies between the input features, resulting in improved segmentation outcomes. However, the high density of attentional feature maps used in this mechanism causes exponential increases in computational complexity. Additionally, it introduces redundant information that negatively impacts the feature representation. Inspired by traditional threshold segmentation algorithms, we propose a novel threshold attention mechanism (TAM). This mechanism significantly reduces computational effort while also better modeling the correlation between different regions of the feature map. Based on TAM, we present a threshold attention network (TANet) for semantic segmentation. TANet consists of an attentional feature enhancement module (AFEM) for global feature enhancement of shallow features and a threshold attention pyramid pooling module (TAPP) for acquiring feature information at different scales for deep features. We have conducted extensive experiments on the ISPRS Vaihingen and Potsdam datasets. The results demonstrate the validity and superiority of our proposed TANet compared to the most state-of-the-art models.

Figures

Figures reproduced from arXiv: 2501.07984 by the authors.

Figure 1
Figure 1. Traditional threshold segmentation method. (a) is the original remote [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. TANet utilizes the ResNet101 backbone network to extract features. Additionally, it employs the AFEM module to enhance the feature information [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. TAM consists of three parts: (a) for thresholding the input features, (b) for calculating the attention weight matrix, and (c) for recovering location [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (5 more)
Figure 5
Figure 5. Figure 5: Structure of the TAPP, where CBR is the convolution layer + BN [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 4
Figure 4. Figure 4: The structure of the AFEM is composed of three branches: the first [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 6
Figure 6. Figure 6: Visualisation of the AFEM module and TAPP module input and output [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Qualitative comparison between our method (TANet) and other methods. The region in the red box represents a challenging segmentation area. [PITH_FULL_IMAGE:figures/full_fig_p010_7.png]
Figure 8
Figure 8. Figure 8: Qualitative comparison between our method (TANet) and other methods. The region in the red box represents a challenging segmentation area. [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 38 canonical work pages

  1. [1]

    Parsing very high resolution urban scene images by learning deep convnets with edge-aware loss,

    X. Zheng, L. Huan, G.-S. Xia, and J. Gong, “Parsing very high resolution urban scene images by learning deep convnets with edge-aware loss,” ISPRS Journal of Photogrammetry and Remote Sensing , vol. 170, pp. 15–28, 2020

    work page 2020

  2. [2]

    Land cover mapping at very high resolution with rotation equivariant cnns: Towards small yet accurate models,

    D. Marcos, M. V olpi, B. Kellenberger, and D. Tuia, “Land cover mapping at very high resolution with rotation equivariant cnns: Towards small yet accurate models,” ISPRS journal of photogrammetry and remote sensing, vol. 145, pp. 96–107, 2018

    work page 2018

  3. [3]

    Rotation- aware and multi-scale convolutional neural network for object detection in remote sensing images,

    K. Fu, Z. Chang, Y . Zhang, G. Xu, K. Zhang, and X. Sun, “Rotation- aware and multi-scale convolutional neural network for object detection in remote sensing images,” ISPRS Journal of Photogrammetry and Remote Sensing, vol. 161, pp. 294–308, 2020

    work page 2020

  4. [4]

    A continuous satellite-derived measure of global terrestrial primary production,

    S. W. Running, R. R. Nemani, F. A. Heinsch, M. Zhao, M. Reeves, and H. Hashimoto, “A continuous satellite-derived measure of global terrestrial primary production,” Bioscience, vol. 54, no. 6, pp. 547–560, 2004

    work page 2004

  5. [5]

    Measuring phenological variability from satellite imagery,

    B. C. Reed, J. F. Brown, D. VanderZee, T. R. Loveland, J. W. Merchant, and D. O. Ohlen, “Measuring phenological variability from satellite imagery,” Journal of vegetation science , vol. 5, no. 5, pp. 703–714, 1994

    work page 1994

  6. [6]

    An aircraft detection framework based on reinforcement learning and convolutional neural networks in remote sensing images,

    Y . Li, K. Fu, H. Sun, and X. Sun, “An aircraft detection framework based on reinforcement learning and convolutional neural networks in remote sensing images,” Remote sensing, vol. 10, no. 2, p. 243, 2018

    work page 2018

  7. [7]

    Detection of collapsed buildings in post-earthquake remote sensing images based on the improved yolov3,

    H. Ma, Y . Liu, Y . Ren, and J. Yu, “Detection of collapsed buildings in post-earthquake remote sensing images based on the improved yolov3,” Remote Sensing, vol. 12, no. 1, p. 44, 2019

    work page 2019

  8. [8]

    Cloud and cloud shadow detection using multilevel feature fused segmentation network,

    Z. Yan, M. Yan, H. Sun, K. Fu, J. Hong, J. Sun, Y . Zhang, and X. Sun, “Cloud and cloud shadow detection using multilevel feature fused segmentation network,” IEEE Geoscience and Remote Sensing Letters, vol. 15, no. 10, pp. 1600–1604, 2018

    work page 2018

Show all 50 references
  1. [9]

    Fully convolutional networks for semantic segmentation,

    J. Long, E. Shelhamer, and T. Darrell, “Fully convolutional networks for semantic segmentation,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2015, pp. 3431–3440

    2015

  2. [10]

    Improving semantic segmentation via decoupled body and edge super- vision,

    X. Li, X. Li, L. Zhang, G. Cheng, J. Shi, Z. Lin, S. Tan, and Y . Tong, “Improving semantic segmentation via decoupled body and edge super- vision,” in European Conference on Computer Vision . Springer, 2020, pp. 435–452

    2020

  3. [11]

    Encoder- decoder with atrous separable convolution for semantic image segmen- tation,

    L.-C. Chen, Y . Zhu, G. Papandreou, F. Schroff, and H. Adam, “Encoder- decoder with atrous separable convolution for semantic image segmen- tation,” in Proceedings of the European conference on computer vision (ECCV), 2018, pp. 801–818

    2018

  4. [12]

    Gated-scnn: Gated shape cnns for semantic segmentation,

    T. Takikawa, D. Acuna, V . Jampani, and S. Fidler, “Gated-scnn: Gated shape cnns for semantic segmentation,” in Proceedings of the IEEE/CVF international conference on computer vision , 2019, pp. 5229–5238

    2019

  5. [13]

    Instance- level human parsing via part grouping network,

    K. Gong, X. Liang, Y . Li, Y . Chen, M. Yang, and L. Lin, “Instance- level human parsing via part grouping network,” in Proceedings of the European conference on computer vision (ECCV) , 2018, pp. 770–785

    2018

  6. [14]

    Afnet: Adaptive fusion network for remote sensing image semantic segmentation,

    R. Liu, L. Mi, and Z. Chen, “Afnet: Adaptive fusion network for remote sensing image semantic segmentation,” IEEE Transactions on Geoscience and Remote Sensing , vol. 59, no. 9, pp. 7871–7886, 2020

    2020

  7. [15]

    Lanet: Local attention embedding to improve the semantic segmentation of remote sensing images,

    L. Ding, H. Tang, and L. Bruzzone, “Lanet: Local attention embedding to improve the semantic segmentation of remote sensing images,” IEEE Transactions on Geoscience and Remote Sensing , vol. 59, no. 1, pp. 426–435, 2020

    2020

  8. [16]

    Aˆ 2-nets: Double attention networks,

    Y . Chen, Y . Kalantidis, J. Li, S. Yan, and J. Feng, “Aˆ 2-nets: Double attention networks,” Advances in neural information processing systems, vol. 31, 2018

    2018

  9. [17]

    Multi-scale context aggregation by dilated convolutions,

    F. Yu and V . Koltun, “Multi-scale context aggregation by dilated convolutions,” arXiv preprint arXiv:1511.07122 , 2015

    2015 arXiv

  10. [18]

    Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs,

    L.-C. Chen, G. Papandreou, I. Kokkinos, K. Murphy, and A. L. Yuille, “Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs,” IEEE transactions on pattern analysis and machine intelligence , vol. 40, no. 4, pp. 834–848, 2017

    2017

  11. [19]

    Hybrid multiple attention network for semantic segmentation in aerial images,

    R. Niu, X. Sun, Y . Tian, W. Diao, K. Chen, and K. Fu, “Hybrid multiple attention network for semantic segmentation in aerial images,” IEEE Transactions on Geoscience and Remote Sensing , vol. 60, pp. 1–18, 2021

    2021

  12. [20]

    Non-local neural net- works,

    X. Wang, R. Girshick, A. Gupta, and K. He, “Non-local neural net- works,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 7794–7803

    2018

  13. [21]

    A deformable attention net- work for high-resolution remote sensing images semantic segmentation,

    R. Zuo, G. Zhang, R. Zhang, and X. Jia, “A deformable attention net- work for high-resolution remote sensing images semantic segmentation,” IEEE Transactions on Geoscience and Remote Sensing , vol. 60, pp. 1– 14, 2021

    2021

  14. [22]

    Learning statistical texture for semantic segmentation,

    L. Zhu, D. Ji, S. Zhu, W. Gan, W. Wu, and J. Yan, “Learning statistical texture for semantic segmentation,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2021, pp. 12 537–12 546

    2021

  15. [23]

    Segnext: Rethinking convolutional attention design for semantic segmentation,

    M.-H. Guo, C.-Z. Lu, Q. Hou, Z.-N. Liu, M.-M. Cheng, and S.-M. Hu, “Segnext: Rethinking convolutional attention design for semantic segmentation,” in NeurIPS, 2022

    2022

  16. [24]

    High-resolution boundary-constrained and context- enhanced network for remote sensing image segmentation,

    Y . Xu and J. Jiang, “High-resolution boundary-constrained and context- enhanced network for remote sensing image segmentation,” Remote Sensing, vol. 14, no. 8, p. 1859, 2022

    2022

  17. [25]

    Ctmfnet: Cnn and transformer multi-scale fusion network of remote sensing urban scene imagery,

    P. Song, J. Li, Z. An, H. Fan, and L. Fan, “Ctmfnet: Cnn and transformer multi-scale fusion network of remote sensing urban scene imagery,” IEEE Transactions on Geoscience and Remote Sensing , 2022

    2022

  18. [26]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, Ł. Kaiser, and I. Polosukhin, “Attention is all you need,” Advances in neural information processing systems , vol. 30, 2017

    2017

  19. [27]

    Ccnet: Criss-cross attention for semantic segmentation,

    Z. Huang, X. Wang, L. Huang, C. Huang, Y . Wei, and W. Liu, “Ccnet: Criss-cross attention for semantic segmentation,” in Proceedings of the IEEE/CVF international conference on computer vision , 2019, pp. 603– 612

    2019

  20. [28]

    Relation-aware global attention for person re-identification,

    Z. Zhang, C. Lan, W. Zeng, X. Jin, and Z. Chen, “Relation-aware global attention for person re-identification,” in Proceedings of the ieee/cvf 12 conference on computer vision and pattern recognition, 2020, pp. 3186– 3195

    2020

  21. [29]

    Spanet: Successive pooling attention network for semantic segmentation of remote sensing images,

    L. Sun, S. Cheng, Y . Zheng, Z. Wu, and J. Zhang, “Spanet: Successive pooling attention network for semantic segmentation of remote sensing images,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , 2022

    2022

  22. [30]

    Beyond self-attention: External attention using two linear layers for visual tasks,

    M.-H. Guo, Z.-N. Liu, T.-J. Mu, and S.-M. Hu, “Beyond self-attention: External attention using two linear layers for visual tasks,” IEEE Transactions on Pattern Analysis and Machine Intelligence , 2022

    2022

  23. [31]

    Squeeze-and-excitation networks,

    J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation networks,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 7132–7141

    2018

  24. [32]

    Cbam: Convolutional block attention module,

    S. Woo, J. Park, J.-Y . Lee, and I. S. Kweon, “Cbam: Convolutional block attention module,” in Proceedings of the European conference on computer vision (ECCV) , 2018, pp. 3–19

    2018

  25. [33]

    Bam: Bottleneck attention module,

    J. Park, S. Woo, J.-Y . Lee, and I. S. Kweon, “Bam: Bottleneck attention module,” arXiv preprint arXiv:1807.06514 , 2018

    2018 arXiv

  26. [34]

    Multiattention network for semantic segmentation of fine-resolution remote sensing images,

    R. Li, S. Zheng, C. Zhang, C. Duan, J. Su, L. Wang, and P. M. Atkinson, “Multiattention network for semantic segmentation of fine-resolution remote sensing images,” IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–13, 2021

    2021

  27. [35]

    Dual attention network for scene segmentation,

    J. Fu, J. Liu, H. Tian, Y . Li, Y . Bao, Z. Fang, and H. Lu, “Dual attention network for scene segmentation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019, pp. 3146– 3154

    2019

  28. [36]

    Rethinking atrous convolution for semantic image segmentation,

    L.-C. Chen, G. Papandreou, F. Schroff, and H. Adam, “Rethinking atrous convolution for semantic image segmentation,” arXiv preprint arXiv:1706.05587, 2017

    2017 arXiv

  29. [37]

    Pyramid scene parsing network,

    H. Zhao, J. Shi, X. Qi, X. Wang, and J. Jia, “Pyramid scene parsing network,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2017, pp. 2881–2890

    2017

  30. [38]

    Ccanet: Class- constraint coarse-to-fine attentional deep network for subdecimeter aerial image semantic segmentation,

    G. Deng, Z. Wu, C. Wang, M. Xu, and Y . Zhong, “Ccanet: Class- constraint coarse-to-fine attentional deep network for subdecimeter aerial image semantic segmentation,” IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–20, 2021

    2021

  31. [39]

    Object-contextual representations for semantic segmentation,

    Y . Yuan, X. Chen, and J. Wang, “Object-contextual representations for semantic segmentation,” in Computer Vision–ECCV 2020: 16th European Conference, Glasgow, UK, August 23–28, 2020, Proceedings, Part VI 16. Springer, 2020, pp. 173–190

    2020

  32. [40]

    Beyond rgb: Very high resolution urban remote sensing with multimodal deep networks,

    N. Audebert, B. Le Saux, and S. Lefèvre, “Beyond rgb: Very high resolution urban remote sensing with multimodal deep networks,” ISPRS journal of photogrammetry and remote sensing , vol. 140, pp. 20–32, 2018

    2018

  33. [41]

    Classification with an edge: Improving semantic image seg- mentation with boundary detection,

    D. Marmanis, K. Schindler, J. D. Wegner, S. Galliani, M. Datcu, and U. Stilla, “Classification with an edge: Improving semantic image seg- mentation with boundary detection,” ISPRS Journal of Photogrammetry and Remote Sensing , vol. 135, pp. 158–172, 2018

    2018

  34. [42]

    Treeunet: Adaptive tree convolutional neural networks for subdecimeter aerial image segmentation,

    K. Yue, L. Yang, R. Li, W. Hu, F. Zhang, and W. Li, “Treeunet: Adaptive tree convolutional neural networks for subdecimeter aerial image segmentation,” ISPRS Journal of Photogrammetry and Remote Sensing, vol. 156, pp. 1–13, 2019

    2019

  35. [43]

    Abcnet: Attentive bilateral contextual network for efficient semantic segmentation of fine-resolution remotely sensed imagery,

    R. Li, S. Zheng, C. Zhang, C. Duan, L. Wang, and P. M. Atkinson, “Abcnet: Attentive bilateral contextual network for efficient semantic segmentation of fine-resolution remotely sensed imagery,”ISPRS journal of photogrammetry and remote sensing , vol. 181, pp. 84–98, 2021

    2021

  36. [44]

    Acfnet: Attentional class feature network for semantic segmentation,

    F. Zhang, Y . Chen, Z. Li, Z. Hong, J. Liu, F. Ma, J. Han, and E. Ding, “Acfnet: Attentional class feature network for semantic segmentation,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2019, pp. 6798–6807

    2019

  37. [45]

    Semantic labeling in very high resolution images via a self-cascaded convolutional neural network,

    Y . Liu, B. Fan, L. Wang, J. Bai, S. Xiang, and C. Pan, “Semantic labeling in very high resolution images via a self-cascaded convolutional neural network,” ISPRS journal of photogrammetry and remote sensing , vol. 145, pp. 78–95, 2018

    2018

  38. [46]

    Semantic segmentation of very-high-resolution remote sensing images via deep multi-feature learning,

    Y . Su, J. Cheng, H. Bai, H. Liu, and C. He, “Semantic segmentation of very-high-resolution remote sensing images via deep multi-feature learning,” Remote Sensing, vol. 14, no. 3, p. 533, 2022

    2022

  39. [47]

    A novel transformer based semantic segmentation scheme for fine-resolution remote sensing images,

    L. Wang, R. Li, C. Duan, C. Zhang, X. Meng, and S. Fang, “A novel transformer based semantic segmentation scheme for fine-resolution remote sensing images,” IEEE Geoscience and Remote Sensing Letters , vol. 19, pp. 1–5, 2022

    2022

  40. [48]

    Dense semantic labeling of subdecimeter reso- lution images with convolutional neural networks,

    M. V olpi and D. Tuia, “Dense semantic labeling of subdecimeter reso- lution images with convolutional neural networks,” IEEE Transactions on Geoscience and Remote Sensing , vol. 55, no. 2, pp. 881–893, 2016

    2016

  41. [49]

    Semantic segmentation of large- size vhr remote sensing images using a two-stage multiscale training architecture,

    L. Ding, J. Zhang, and L. Bruzzone, “Semantic segmentation of large- size vhr remote sensing images using a two-stage multiscale training architecture,” IEEE Transactions on Geoscience and Remote Sensing , vol. 58, no. 8, pp. 5367–5376, 2020

    2020

  42. [50]

    Problems of encoder-decoder frame- works for high-resolution remote sensing image segmentation: Structural stereotype and insufficient learning,

    Y . Sun, Y . Tian, and Y . Xu, “Problems of encoder-decoder frame- works for high-resolution remote sensing image segmentation: Structural stereotype and insufficient learning,”Neurocomputing, vol. 330, pp. 297– 304, 2019

    2019

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.