Pith. sign in

REVIEW 3 major objections 6 minor 72 references

A Novel Shape Guided Transformer Network for Instance Segmentation in Remote Sensing Images

T0 review · 3 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 a two-stage instance segmentation network, SGTN, combining a long-range-correlation Swin-style encoder (LSwin) and a shape guidance module, achieves the highest mask average precision on the WHU, BITCC, and NWPU…

desk verdict Useful applied architecture for RSI instance segmentation with consistent ablations, but the SOTA claim rests on a 0.2 AP margin over in-house reimplemented baselines with no error bars. read the letter →

arxiv 2501.00360 v1 pith:ULAVNSTY submitted 2024-12-31 cs.CV cs.LG

Dawen Yu , Shunping Ji This is my paper · ORCID
classification cs.CVcs.LG
keywords instancesegmentationremotesensingimagestransformerencoderlong-rangecorrelationshapeguidanceself-attentionbuildingaverageprecision
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 is trying to establish that instance segmentation in remote sensing images can be improved by attacking two bottlenecks at once: relating instances scattered across large scenes, and recovering their boundaries and shapes under atmospheric degradation. The proposed SGTN combines a transformer encoder called LSwin with a shape guidance module; LSwin adds vertical and horizontal one-dimensional global self-attention to the local shifted-window Swin Transformer, and the shape module supervises class-agnostic foreground, edge, and corner maps and then uses the foreground map to refine predicted instance masks. The paper reports the best average precision scores on the WHU, BITCC, and NWPU VHR-10 datasets against all compared methods, with ablations giving consistent gains to both components. A sympathetic reader would care because the ingredients are inexpensive: the added attention is linear, and the shape maps come from simple supervised heads rather than a new segmentation paradigm.

What carries the argument

The load-bearing objects are the LSwin block and the shape guidance module. A LSwin block interleaves two ordinary Swin Transformer blocks with two long-range correlation blocks; the correlation blocks perform self-attention along full rows (H-MSA) and full columns (V-MSA) instead of within local 2-D windows, with instance normalization inside each 1-D window. The two streams are fused as $\alpha z_{sw} + \beta z_{lr}$, with $\alpha=1$ and $\beta=0$ initially so that the pretrained Swin weights dominate at the start. The shape guidance module starts from a shallow detail-rich feature, predicts class-agnostic foreground, edge, and corner maps under three binary cross-entropy losses (edge pixels doubled, corner pixels quadrupled in weight), and then refines each resized instance mask by the Hadamard product with the cropped foreground map, $M_i = M_c \odot M_s$. These mechanisms carry the argument: one supplies global context at linear cost, the other supplies boundary-preserving detail.

What would settle it

Re-run the same baselines with the authors' training setup and multiple random seeds, or with their published official code and checkpoints; if CenterMask or DANCE matches or exceeds SGTN's average precision on WHU, BITCC, or NWPU VHR-10 once noise is accounted for, the claimed superiority is not established. A narrower test is to ablate Swin-S versus LSwin inside SGTN over several seeds and check whether the reported 0.5 to 2.2 average-precision gaps exceed the run-to-run spread.

Watch

Extended reading notes

Core claim

The central claim is that local detail and global context are complementary and can be fused in one end-to-end network. LSwin is built on Swin-S by interleaving long-range correlation blocks with the original shifted-window blocks; each long-range block runs self-attention along rows (H-MSA) and along columns (V-MSA) with instance normalization, giving explicit cross-image interaction at linear cost, and the two block streams are combined with learnable coefficients $\alpha$ and $\beta$ initialized to $1$ and $0$. In parallel, the shape guidance module concatenates a shallow detail-rich feature with the encoder feature, supervises three binary maps (foreground, edges, corners) with weighted binary cross-entropy, and refines each RoI-aligned mask by element-wise multiplication with the predicted foreground map, $M_i = M_c \odot M_s$. The paper claims this combination yields the highest mask AP on two single-class building datasets and one ten-class dataset, with LSwin improving over ResNet-50 and Swin-S at similar efficiency and with SGM adding 1.1 to 1.4 AP points in ablations.

Load-bearing premise

The headline ranking rests on the authors' re-implementations of every baseline being at least as carefully tuned as their own method, since several winning margins are under half a percentage point of average precision and no repeated runs or error bars are reported.

Editorial extensions

If this is right

  • LSwin is a drop-in encoder: the paper shows it raises average precision for CenterMask and DANCE as well as for SGTN, so the global-attention design transfers to other instance segmentation frameworks.
  • The two attention paradigms are complementary: LSwin beats both Swin-S and a pure long-range-correlation encoder on all three datasets, which implies that local window attention and 1-D global attention are not substitutes.
  • SGM's gains concentrate at high IoU thresholds (AP75 up 0.8, 0.9, and 2.1 percentage points), so the shape module is improving boundary quality rather than merely adding detections.
  • With the same ResNet-50 backbone, SGTN still beats the second-best CenterMask by 0.2, 1.0, and 2.8 AP on the three datasets, so the reported advantage is not solely an effect of the transformer backbone.

Reading between the lines

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

  • One untested consequence of the linear attention design is that LSwin's relative advantage over Swin should grow as tile size grows, because 1-D global attention scales linearly while the number of windows grows; running the same comparison on larger crops would test this directly.
  • Because the shape supervision is category-agnostic, the foreground/edge/corner maps may transfer across datasets and even to new object classes without retraining; the paper does not report such cross-dataset experiments.
  • The Hadamard fusion rule is a hard gate: it can only remove pixels from the resized mask, never add them. A learned or soft fusion could reveal how much of SGM's gain comes from the gate versus the auxiliary supervision itself.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper proposes SGTN, a two-stage instance segmentation network for remote sensing images, combining a new transformer encoder LSwin and a shape guidance module (SGM). LSwin augments Swin-S with vertical and horizontal 1D global self-attention blocks, with learnable fusion weights α and β. SGM produces category-agnostic foreground, edge, and corner predictions from a fused shallow-and-deep feature, and its foreground map is used to refine predicted instance masks via a Hadamard product. The authors report state-of-the-art AP on WHU, BITCC, and NWPU VHR-10, and present ablations showing consistent gains from both LSwin and SGM across all three datasets.

Significance. If the empirical claims are reliable, the combination of local shifted-window attention with 1D global self-attention is a practical way to improve global context modeling for remote sensing instance segmentation without the full cost of 2D global attention, and the SGM's foreground-refinement mechanism is a simple and transferable module. The ablations are systematic and consistent across three datasets, which lends internal support to the contribution of each component. However, the significance is currently limited by the fragility of the headline results: the main SOTA margin on the WHU dataset is 0.2 AP against an in-house reimplementation of CenterMask, no variance or significance information is provided, the runtime claim for LSwin is contradicted by Table V, and no code or official baseline checkpoints are made available to reproduce the comparison.

major comments (3)
  1. [Section IV-B.2, Tables I–III] The central claim of achieving the highest AP on the WHU dataset rests on a 0.2 AP margin over an in-house reimplementation of CenterMask (72.4 vs 72.2, Table I), and no standard deviations, repeated runs, or significance tests are reported for any result. The WHU ablation in Table IV attributes 1.3 AP to the SGM alone, so the entire cross-method winning margin is smaller than the effect of a single component; this makes the SOTA claim vulnerable to run-to-run noise, data augmentation differences, or slight under-tuning of the reimplemented baselines. Please provide multiple-seed runs with variance estimates for at least the main comparisons, and either use official released code or checkpoints for all baselines or justify in detail that the in-house implementations reproduce the published configurations.
  2. [Section IV-F.1 and Table V] The abstract and Section IV-D state that LSwin is better than ResNet and Swin at the same level of efficiency, but Table V reports 66.3 sec for SGTN with LSwin versus 52.6 sec with ResNet-50 and 56.0 sec with Swin-S on the BITCC dataset, which is roughly a 26% increase over the ResNet-50 variant. The text that describes these numbers as 'almost equally efficient' is an overstatement. Please either correct the efficiency claim to reflect the measured runtime differences or provide an explanation of the throughput conditions under which the claim holds.
  3. [Section III-A and Table IV] The mask refinement step Mi = Mc ⊙ Ms is presented as a key use of the SGM, and the ablation in Table IV measures the effect of the whole SGM including the auxiliary losses and the fusion rule together. Since the whole-module gain of 1.3 AP on WHU exceeds the 0.2 AP SOTA margin, the paper should isolate whether the improvement comes from the multi-label shape supervision, the Hadamard-product refinement, or the simple concatenation of shallow and deep features. An additional ablation removing only the mask-refinement step (keeping the auxiliary losses) would clarify this and make the design justification load-bearing rather than implicit.
minor comments (6)
  1. [Section II-C and III-B] The 1D vertical/horizontal self-attention idea is introduced as a 'new paradigm' and credited only to the authors' previous work [8], but similar axial/1D attention schemes have been used in vision transformers; the related work discussion should acknowledge this prior art and position LSwin as an integration rather than a wholly new mechanism.
  2. [Section III-B, Eqs. (1)–(2)] The notation in equations (1) and (2) is inconsistent (for example, z^l_sw and z_lr^l+1 are used without a shared convention for layer indexing), which makes the block equations difficult to follow; please unify the superscript and subscript notation.
  3. [Section IV-C.1] The text says 'YOLCAT' instead of 'YOLACT' in the WHU experiment description, and the reference numbers in Table II do not match the bibliography order (e.g., YOLACT is cited as [13] instead of [14]); these should be corrected.
  4. [Fig. 13 caption] The caption reads 'from SG-LRN', which appears to be a leftover from a previous draft and should be replaced with 'from SGTN'.
  5. [Section IV-F.2] The subsection title says 'Long-Range Correction blocks', but the term used elsewhere is 'Long-Range Correlation'; please make the terminology consistent.
  6. [Section IV-B.2] The manuscript promises 'Code will be available' but no code or model checkpoints are provided; given that all baselines were reimplemented in-house, public code would be particularly important for verifying the comparison.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: SGTN's claims are empirical, benchmarked against external datasets, and its LSwin and SGM components are validated by ablations rather than reducing to their own inputs.

full rationale

The paper contains no mathematical derivation that could collapse into its inputs. Its central claims are empirical: SGTN achieves the highest AP on the WHU, BITCC, and NWPU VHR-10 public benchmarks. These benchmarks and most baselines are external, and the comparison methods are re-implemented from published architectures. The self-citations [8] and [49] provide a 1D self-attention mechanism and an anchor-free candidate box generation module, but they are not load-bearing in a circular way: LSwin is ablated against Swin-S and LRC (Tables I-III and VI), and the SGM is ablated in Table IV. No fitted parameter is renamed as a prediction, no uniqueness theorem is imported from the authors' prior work, and no result is defined in terms of the claim it is supposed to support. The only substantive concern is the fairness and statistical stability of the comparison, especially the 0.2 AP margin over the in-house CenterMask reimplementation on WHU with no error bars or significance tests. That is a correctness and reproducibility risk, not a circularity risk. Therefore, no circular step can be identified from the text.

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

The central claim is an empirical performance claim. It rests on the choice of public benchmarks, the in-house reimplementation of baselines, unstated generation of edge and corner ground truth, and the assumption that single-run AP differences of a few tenths of a point are meaningful. No free parameters beyond standard training hyperparameters are fitted to the test sets, and no invented physical entities are postulated.

free parameters (3)
  • Fusion coefficients alpha and beta = learned; initialized alpha=1, beta=0
    Section III-B: adaptive fusion of Swin and long-range correlation block outputs; initial values chosen so the pretrained Swin path dominates at the start of training.
  • Edge and corner loss weights = 2x edge, 4x corner
    Section III-E: weight coefficients doubled and quadrupled for edge and corner pixels in all three BCE losses; hand-set without reported tuning.
  • Training epochs per dataset = 30 (WHU), 60 (BITCC), 100 (NWPU VHR-10)
    Section IV-B.2: set based on sample scales; no tuning procedure reported.
assumptions (5)
  • domain assumption Fair re-implementation of all comparison methods
    Section IV-B.2 states all comparison methods were re-implemented with the same PyTorch framework and augmentation; no official checkpoints or published numbers are used.
  • domain assumption Single-run AP differences are meaningful
    Tables I-IV report one set of AP numbers without variance; gains as small as 0.2 to 0.5 percentage points are treated as significant.
  • domain assumption Ground-truth edge and corner maps can be produced from instance masks
    Section III-E: SGM supervision needs pixel-wise maps for foreground, edges, and corners, but the generation procedure is not described.
  • domain assumption Row and column 1D attention captures needed global context
    Section III-C: the design assumes vertical and horizontal correlations provide sufficient long-range dependency for RSI segmentation; supported only by ablations on three datasets.
  • domain assumption Pretrained Swin-S weights transfer to remote sensing imagery
    Section III-B: Swin Transformer blocks are initialized with ImageNet-pretrained weights; this is standard but not examined in the paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Novel Shape Guided Transformer Network for Instance Segmentation in Remote Sensing Images." pith.science (2026). https://pith.science/paper/ULAVNSTY

@misc{pith2026250100360,
  author       = {Pith},
  title        = {Pith review of: A Novel Shape Guided Transformer Network for Instance Segmentation in Remote Sensing Images},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ULAVNSTY}},
  note         = {Machine review of arXiv:2501.00360}
}
read the original abstract

Instance segmentation performance in remote sensing images (RSIs) is significantly affected by two issues: how to extract accurate boundaries of objects from remote imaging through the dynamic atmosphere, and how to integrate the mutual information of related object instances scattered over a vast spatial region. In this study, we propose a novel Shape Guided Transformer Network (SGTN) to accurately extract objects at the instance level. Inspired by the global contextual modeling capacity of the self-attention mechanism, we propose an effective transformer encoder termed LSwin, which incorporates vertical and horizontal 1D global self-attention mechanisms to obtain better global-perception capacity for RSIs than the popular local-shifted-window based Swin Transformer. To achieve accurate instance mask segmentation, we introduce a shape guidance module (SGM) to emphasize the object boundary and shape information. The combination of SGM, which emphasizes the local detail information, and LSwin, which focuses on the global context relationships, achieve excellent RSI instance segmentation. Their effectiveness was validated through comprehensive ablation experiments. Especially, LSwin is proved better than the popular ResNet and Swin transformer encoder at the same level of efficiency. Compared to other instance segmentation methods, our SGTN achieves the highest average precision (AP) scores on two single-class public datasets (WHU dataset and BITCC dataset) and a multi-class public dataset (NWPU VHR-10 dataset). Code will be available at http://gpcv.whu.edu.cn/data/.

Figures

Figures reproduced from arXiv: 2501.00360 by the authors.

Figure 1
Figure 1. The proposed instance segmentation method incorporates two key components that achieve local detail perception and global context perception, respectively. Component A is the new transformer encoder, and component B is the shape guidance module. They are respectively used to tackle the long￾term dependency and precise boundary recognition problem in RSI instance segmentation tasks, aiming to achieve accurate instanc… view at source ↗
Figure 2
Figure 2. Fig.2. The network structure of our proposed SGTN. Candidate box [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. The architecture of the Long-range correlation boosted Swin [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: The architecture of LSwin block. A shows two successive [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 6
Figure 6. Figure 6: Illustration of the self-attention operation in H-MSA [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 8
Figure 8. Figure 8: Examples of remote sensing images and their corresponding [PITH_FULL_IMAGE:figures/full_fig_p006_8.png]
Figure 9
Figure 9. Figure 9: The building instance segmentation results on the WHU data [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]
Figure 10
Figure 10. Figure 10: The building instance segmentation results on the BITCC dataset from different methods. The predicted true positive, false positive, and false negative instances in the results are colored in blue, red, and orange, respectively. (a)~(g) are the results from ResNet-50 …
Figure 11
Figure 11. Figure 11: The performance comparison of different methods on the NW [PITH_FULL_IMAGE:figures/full_fig_p010_11.png]
Figure 12
Figure 12. Figure 12: The performance comparison between different instance segmentation [PITH_FULL_IMAGE:figures/full_fig_p010_12.png]
Figure 13
Figure 13. Figure 13: The instance segmentation results on the NWPU VHR-10 dataset from SG-LRN. The predicted true positive instances are annotated with randomly colored bounding boxes attached with their category texts. The predicted false negative and false positive instances are annotat…
Figure 14
Figure 14. Figure 14: The performance comparison of three different feature en [PITH_FULL_IMAGE:figures/full_fig_p011_14.png]
Figure 15
Figure 15. Figure 15: Visual comparison of prediction results with and without embedding SGM in the proposed method. Table VI COMPARISON EXPERIMENTS FOR DIFFERENT ENCODERS. Dataset Backbone AP (%) AP50 (%) AP75 (%) WHU dataset Swin-S 74.1 92.7 84.4 LRC 71.2 90.7 81.2 LSwin 74.6 93.2 84.7 B…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

72 extracted references · 69 canonical work pages

  1. [1]

    We propose SGTN, a novel end-to-end model for RSI instance segmentation, which focuses on the two key issues, the long-term dependency and precise boundary recognition by incorporating a new transformer encoder and a shape guidance module

    work page

  2. [2]

    The developed encoder offers superior global perception capability for large-scale RSIs, efficiently integrating mutual information to interpret related object instances scattered across broad spatial regions. The designed shape guidance module enhances boundary perception through specialized supervision, generating shape-preserving features and improving...

    work page

  3. [3]

    The NWPU VHR-10 dataset contains 650 remote sensing images and corresponding instance annotations collected from Google Earth and ISPRS Vaihingen datasets [28]

    NWPU VHR-10 dataset. The NWPU VHR-10 dataset contains 650 remote sensing images and corresponding instance annotations collected from Google Earth and ISPRS Vaihingen datasets [28]. The image size ranges from 430×543 pixels to 1028×1728 pixels, with a ground resolution of 0.08-2 meters. A total of 10 categories were labeled, including airplanes (c1), ship...

    work page

  4. [4]

    W, H and C

    proposed an online hard sample mining strategy to enhance the network's attention to object edges, they assigned higher weights to pixels closer to the hard-to-segment instance boundaries. Wang et al. [1] explicitly modeled feature representations for the edges of objects by employing additional edge supervision. In contrast to the above methods that only...

    work page

  5. [5]

    WHU dataset. The WHU dataset is a large-scale remote sensing building instance segmentation benchmark dataset [53], which consists of high-resolution aerial images and around 186,500 building instance annotations, collected from Christchurch, New Zealand. The dataset is divided into a training set (70% of the buildings), a validation set (10% of the build...

    work page

  6. [6]

    BITCC dataset. The Building Instances of Typical Cities in China dataset (BITCC dataset for short) [54] consists of high- resolution satellite images and the corresponding building instance annotations, collected from Beijing, Shanghai, Wuhan, and Shenzhen, China. The size of original images in the BITCC dataset is 500×500 pixels. We padded images at the ...

    work page

  7. [7]

    Multi-Swin Mask Transformer for Instance Segmentation of Agricultural Field Extraction,

    B. Zhong, T. Wei, X. Luo et al., “Multi-Swin Mask Transformer for Instance Segmentation of Agricultural Field Extraction,” Remote Sens., vol. 15, no. 3, pp. 549, 2023

    work page 2023

  8. [8]

    Evaluation Metrics: We evaluate the performance of different methods with the standard MS COCO measure [55]. The average precision (AP) at 10 different mask intersection over union (IoU) thresholds for instances of all sizes is taken as the main criterion, which can be computed as follows:  10 50 1 51 10 i ii AP AP    (5) where APi is the average...

    work page

Show all 72 references
  1. [9]

    and the more recent Swin Transformer [10] have achieved notable success in the computer vision field. However, ViT is limited by high computational requirements and low spatial resolution for feature extraction; Swin Transformer has certain limitations when applied to RSIs due...

  2. [10]

    Implementation details: All experiments in this paper were conducted on a Windows PC equipped with an NVIDIA GeForce RTX 3090 24G GPU and an Intel Core i9-12900KF CPU. For a fair comparison, we re-implemented all the comparison methods, including the one- stage methods YOLACT ...

  3. [11]

    The quantitative results of the proposed SGTN and other comparison methods are listed in Table I

    WHU dataset. The quantitative results of the proposed SGTN and other comparison methods are listed in Table I. The comparison methods, i.e., YOLCAT, SOLO, Mask R-CNN, Deep Snake, Center-Mask, DANCE, and Buildmapper, all use ResNet-50 as their backbone. For the proposed SGTN, w...

  4. [12]

    The quantitative results from different instance segmentation methods on the BITCC dataset are listed in Table II

    BITCC dataset. The quantitative results from different instance segmentation methods on the BITCC dataset are listed in Table II. Similar to the conclusions drawn from the WHU dataset, LSwin is highly effective for remote sensing instance segmentation tasks, providing signific...

  5. [13]

    Results on the multi-category NWPU VHR-10 dataset from different instance segmentation methods are summarized in Table III

    NWPU VHR-10 dataset. Results on the multi-category NWPU VHR-10 dataset from different instance segmentation methods are summarized in Table III. SGTN achieves an AP of 69.3%, surpassing the second-best CenterMask by 2.8% when using ResNet-50 as the backbone. This highlights SG...

    2019

  6. [14]

    In this subsection, we analyze the inference efficiency of our newly developed method for instance segmentation of remote sensing images

    Inference Efficiency. In this subsection, we analyze the inference efficiency of our newly developed method for instance segmentation of remote sensing images. We conducted this analysis using the multi-category BITCC dataset, and the runtime statistics of all comparison metho...

  7. [15]

    W/” DENOTES WITH, “W/O

    Building an encoder from only long-range correction blocks. In this work, we build a new encoder by combining Swin Transformer blocks and efficient 1D Long-Range Correction (LRC) blocks. In this subsection, we discuss the effectiveness of using single LRC blocks within the enc...

  8. [16]

    Mask Decoupled Head for Instance Segmentation in Remote Sensing Images

    Y. Wang, X. Zhang, T. Zhang et al., "Mask Decoupled Head for Instance Segmentation in Remote Sensing Images." in Proc. IEEE Int. Geosci. Remote Sens. Symp. (IGARSS), 2022, pp. 3123-3126

    2022

  9. [17]

    Instance Segmentation in Very High Resolution Remote Sensing Imag ery Based on Hard-to-Segment Instance Learning and Boundary Shape Analysis,

    Y. Gong, F. Zhang, X. Jia et al., “Instance Segmentation in Very High Resolution Remote Sensing Imag ery Based on Hard-to-Segment Instance Learning and Boundary Shape Analysis,” Remote Sens., vol. 14, no. 1, pp. 23, 2021

    2021

  10. [18]

    Accurate Instance Segmentation for Remote Sensing Images via Adaptive and Dynamic Feature Learning,

    F. Yang, X. Yuan, J. Ran et al., “Accurate Instance Segmentation for Remote Sensing Images via Adaptive and Dynamic Feature Learning,” Remote Sens., vol. 13, no. 23, pp. 4774, 2021

    2021

  11. [19]

    Simultaneous detection and segmentation

    B. Hariharan, P. Arbeláez, R. Girshick et al., "Simultaneous detection and segmentation." in Proc. Eur. Conf. Comput. Vis. (ECCV), 2014, pp. 297-312

    2014

  12. [20]

    An anchor-free network with box refinement and saliency supplement for instance segmentation in remote sensing images,

    F. Shi, and T. Zhang, “An anchor-free network with box refinement and saliency supplement for instance segmentation in remote sensing images,” IEEE Geosci. Remote Sens. Lett., vol. 19, pp. 1-5, 2022

    2022

  13. [21]

    Bounding box-free instance segmentation using semi-supervised iterative learning for vehicle detection

    O. L. F. de Carvalho et al., "Bounding box-free instance segmentation using semi-supervised iterative learning for vehicle detection", IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens. , vol. 15, pp. 3403-3420, 2022

    2022

  14. [22]

    Polarmask: Single shot instance segmentation with pola r representation

    E. Xie, P. Sun, X. Song et al., "Polarmask: Single shot instance segmentation with pola r representation." in Proc. IEEE Int. Conf. Comput. Vis. Pattern Recog., 2020, pp. 12193-12202

    2020

  15. [23]

    Long-Range Correlation Supervision for Land-Cover Classification from Remote Sensing Images,

    D. Yu, S. Ji, “Long-Range Correlation Supervision for Land-Cover Classification from Remote Sensing Images,” IEEE Trans. Geosci. Remote Sens., vol. 61, pp. 1-14, 2023

    2023

  16. [24]

    An image is worth 16x16 words: Transformers for image recognition at scale,

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, and N. Houlsby, “An image is worth 16x16 words: Transformers for image recognition at scale,” in 9th International Conference on Lear...

    2021

  17. [25]

    Swin transformer: Hierarchical vision transformer using shifted windows

    Z. Liu, Y. Lin, Y. Cao et al., "Swin transformer: Hierarchical vision transformer using shifted windows." in Proc. IEEE Int. Conf. Comput. Vis., 2021, pp. 10012-10022

    2021

  18. [26]

    An improved swin transformer-based model for remote sensing object detection and instance segmentation,

    X. Xu, Z. Feng, C. Cao et al., “An improved swin transformer-based model for remote sensing object detection and instance segmentation,” Remote Sens., vol. 13, no. 23, pp. 4779, 2021

    2021

  19. [27]

    RSPrompter: Learning to prompt for remote sensing instance segmentation based on visual foundation model,

    K. Chen, C. Liu, H. Chen et al., “RSPrompter: Learning to prompt for remote sensing instance segmentation based on visual foundation model,” IEEE Trans. Geosci. Remote Sens., vol. 62, pp. 1-17, 2024

    2024

  20. [28]

    Contour Loss: Boundary-Aware Learning for Salient Object Segmentation,

    Z. Chen, H. Zhou, X. Xie et al., “Contour Loss: Boundary-Aware Learning for Salient Object Segmentation,” Unpublished paper, 2019. [Online]. Available: https://arxiv.org/abs/1908.01975

    2019 arXiv

  21. [29]

    Yolact: Real-time instance segmentation

    D. Bolya, C. Zhou, F. Xiao et al., "Yolact: Real-time instance segmentation." in Proc. IEEE Int. Conf. Comput. Vis. , 2019, pp. 9157- 9166

    2019

  22. [30]

    Solo: Segmenting objects by locations,

    X. Wang, T. Kong, C. Shen et al., "Solo: Segmenting objects by locations," in Proc. Eur. Conf. Comput. Vis., 2020, pp. 649-665

    2020

  23. [31]

    Blendmask: Top-down meets bottom- up for instance segmentation

    H. Chen, K. Sun, Z. Tian et al., "Blendmask: Top-down meets bottom- up for instance segmentation." in Proc. IEEE Int. Conf. Comput. Vis. Pattern Recog., 2020, pp. 8573-8581

    2020

  24. [32]

    Conditional convolutions for instance segmentation

    Z. Tian, C. Shen, and H. Chen, "Conditional convolutions for instance segmentation." in Proc. Eur. Conf. Comput. Vis. (ECCV), 2020, pp. 282- 298

    2020

  25. [33]

    Mask R-CNN,

    K. He, G. Gkioxari, P. Dollár et al., "Mask R-CNN," in Proc. IEEE Int. Conf. Comput. Vision.,2017, pp. 2961-2969

    2017

  26. [34]

    Faster R-CNN: Towards Real- Time Object Detection with Region Proposal Networks,

    S. Ren, K. He, R. Girshick and J. Sun, "Faster R-CNN: Towards Real- Time Object Detection with Region Proposal Networks," IEEE Trans. Pattern Anal. Mach. Intell., vol. 39, pp. 1137-1149, 2017

    2017

  27. [35]

    Hybrid task cascade for instance segmentation,

    K. Chen, J. Pang, J. Wang et al., "Hybrid task cascade for instance segmentation," in Proc. IEEE Int. Conf. Comput. Vis. Pattern Recog. , 2019, pp. 4974-4983

    2019

  28. [36]

    Path aggregation network for instance segmentation,

    S. Liu, L. Qi, H. Qin et al., "Path aggregation network for instance segmentation," in Proc. IEEE Int. Conf. Comput. Vis. Pattern Recog. , 2018, pp. 8759-8768

    2018

  29. [37]

    Global contex t parallel attentio n for anchor-free instance segmentation in remote sensing images,

    X. Liu, and X. Di, “Global contex t parallel attentio n for anchor-free instance segmentation in remote sensing images,” IEEE Geosci. Remote Sens. Lett., vol. 19, pp. 1-5, 2020

    2020

  30. [38]

    Fas t interactive object annotation with curve-gcn,

    H. Ling, J. Gao, A. Kar et al., "Fas t interactive object annotation with curve-gcn," in Proc. IEEE Int. Conf. Comput. Vis. Pattern Recog., 2019, pp. 5257-5266

    2019

  31. [39]

    Deep snake for real-time instance segmentation,

    S. Peng, W. Jiang, H. Pi et al., "Deep snake for real-time instance segmentation," in Proc. IEEE Int. Conf. Comput. Vis. Pattern Recog. , 2020, pp. 8533-8542

    2020

  32. [40]

    Dance: A deep attentive contour model for efficient instance segmentation,

    Z. Liu, J. H. Liew, X. Chen et al., "Dance: A deep attentive contour model for efficient instance segmentation," in Proc. IEEE Winter Conf. Appl. Comput. Vis. (WACV), 2021, pp. 345-354

    2021

  33. [41]

    Annotating object instances with a polygon-rnn

    L. Castrejon, K. Kundu, R. Urtasun et al., "Annotating object instances with a polygon-rnn." in Proc. IEEE Int. Conf. Comput. Vis. Pattern Recog., 2017, pp. 5230-5238

    2017

  34. [42]

    Efficient interactive annotation of segmentation datasets with polygon-rnn++

    D. Acuna, H. Ling, A. Kar et al., "Efficient interactive annotation of segmentation datasets with polygon-rnn++." in Proc. IEEE Int. Conf. Comput. Vis. Pattern Recog., 2018, pp. 859-868

    2018

  35. [43]

    HQ-ISNet: High-quality instance segmentation for remote sensing imagery,

    H. Su, S. Wei, S. Liu et al., “HQ-ISNet: High-quality instance segmentation for remote sensing imagery,” Remote Sens., vol. 12, no. 6, pp. 989, 2020

    2020

  36. [44]

    Ship instance segmentation from remote sensing images using sequence local context module

    Y. Feng, W. Diao, Z. Chang, M. Yan, X. Sun and X. Gao, "Ship instance segmentation from remote sensing images using sequence local context module", in Proc. IEEE Int. Geosci. Remote Sens. Symp. (IGARSS), 2019, pp. 1025-1028

    2019

  37. [45]

    DB-BlendMask: Decomposed attention and balanced BlendMask fo r instance segmentation of high- resolution remote sensing images,

    Z. Chen, Y. Shang, A. Python et al., “DB-BlendMask: Decomposed attention and balanced BlendMask fo r instance segmentation of high- resolution remote sensing images,” IEEE Trans. Geosci. Remote Sens. , vol. 60, pp. 1-15, 2021

    2021

  38. [46]

    Faster and Better Instance Segmentation for Large Scene Remote Sensing Imagery

    H. Su, P. Huang, J. Yin et al., "Faster and Better Instance Segmentation for Large Scene Remote Sensing Imagery." in Proc. IEEE Int. Geosci. Remote Sens. Symp. (IGARSS), 2022, pp. 2187-2190

    2022

  39. [47]

    Precise and robust ship detection for high- resolution SAR imagery based on HR-SDNet,

    S. Wei, H. Su, J. Ming et al., “Precise and robust ship detection for high- resolution SAR imagery based on HR-SDNet,” Remote Sens., vol. 12, no. 1, pp. 167, 2020

    2020

  40. [48]

    Cas cade R-CNN: High quality object detection and instance segmentation,

    Z. Cai, and N. Vasconcelos, “Cas cade R-CNN: High quality object detection and instance segmentation,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 43, no. 5, pp. 1483-1498, 2019

    2019

  41. [49]

    OEC-RNN: Object-oriented delineation of rooftops with edges and corners using the recurrent neural network from the aerial images,

    W. Huang, H. Tang, and P. Xu, “OEC-RNN: Object-oriented delineation of rooftops with edges and corners using the recurrent neural network from the aerial images,” IEEE Trans. Geosci. Remote Sens., vol. 60, pp. 1-12, 2021

    2021

  42. [50]

    Building outline delineation: From aerial images to polygons with an improved end-to-end learning framework,

    W. Zhao, C. Persello, and A. Stein, “Building outline delineation: From aerial images to polygons with an improved end-to-end learning framework,” ISPRS J. Photogramm. Remote Sens. , vol. 175, pp. 119- 131, 2021

    2021

  43. [51]

    BuildMapper: A fully learnable framework for vectorized building contour extraction,

    S. Wei, T. Zhang, S. Ji et al ., “BuildMapper: A fully learnable framework for vectorized building contour extraction,” ISPRS J. Photogramm. Remote Sens., vol. 197, pp. 87-104, 2023

    2023

  44. [52]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar et al., “Attention is all you need,” Adv. neural inf. proces. syst., vol. 30, pp. 5998–6008, 2017

    2017

  45. [53]

    Learning to aggregate multi-scale context for instance segmentation in remote sensing images,

    Y. Liu, H. Li, C. Hu et al., “Learning to aggregate multi-scale context for instance segmentation in remote sensing images,” IEEE Trans. Neural Netw. Learn. Syst., pp. 1-15, 2024

    2024

  46. [54]

    GLSANet: Global-local self-attention network for remote sensing image semantic segmentation,

    X. Hu, P. Zhang, Q. Zhang et al., “GLSANet: Global-local self-attention network for remote sensing image semantic segmentation,” IEEE Geosci. Remote Sens. Lett., vol. 20, pp. 1-5, 2023

    2023

  47. [55]

    LPASS-Net : Lightweight progressive attention semantic segmentation network for automatic segmentation of remote sensing images,

    H. Liang, and S. Seo, “LPASS-Net : Lightweight progressive attention semantic segmentation network for automatic segmentation of remote sensing images,” Remote Sens., vol. 14, no. 23, pp. 6057, 2022

    2022

  48. [56]

    Swin Transformer Embedding UNet for Remote Sensing Image Semantic Segmentation,

    X. He, Y. Zhou, J. Zhao et al., “Swin Transformer Embedding UNet for Remote Sensing Image Semantic Segmentation,” IEEE Trans. Geosci. Remote. Sens., vol. 60, pp. 1-15, 2022. > REPLACE THIS LINE WITH YOUR PAPER IDENTIFICATION NUMBER (DOUBLE-CLICK HERE TO EDIT) < 14

    2022

  49. [57]

    Swin-transformer-enabled YOLOv5 with attention mechanism for small objec t detection on sa tellite images,

    H. Gong, T. Mu, Q. Li et al., “Swin-transformer-enabled YOLOv5 with attention mechanism for small objec t detection on sa tellite images,” Remote Sens., vol. 14, no. 12, pp. 2861, 2022

    2022

  50. [58]

    FN" and

    (2024) were evaluated. Notably, when using the same backbone, SGTN consistently achieves higher AP scores compared to other methods. Moreover, LSwin, developed in this study, exhibits significant advantages when compared across different backbones. Fig. 12 clearly illustrates ...

    2024

  51. [59]

    Segment anything

    A. Kirillov, E. Mintun, N. Ravi et al., "Segment anything." in Proc. IEEE Int. Conf. Comput. Vis.,2023, pp. 4015-4026

    2023

  52. [60]

    Gated-scnn: Gated shape cnns for semantic segmentation

    T. Takikawa, D. Acuna, V. Jampani et al., "Gated-scnn: Gated shape cnns for semantic segmentation." in Proc. IEEE Int. Conf. Comput. Vision., 2019, pp. 5229-5238

    2019

  53. [61]

    Semantic image segmentation with task-specific edge detection using cnns and a discriminatively trained domain transform

    L.-C. Chen, J. T. Barron, G. Papandreou et al., "Semantic image segmentation with task-specific edge detection using cnns and a discriminatively trained domain transform." In Proc. IEEE Int. Conf. Comput. Vis. Pattern Recog, 2016, pp. 4545-4554

    2016

  54. [62]

    Boundary-preserving mask r-cnn

    T. Cheng, X. Wang, L. Huang et al., "Boundary-preserving mask r-cnn." in Proc. Eur. Conf. Comput. Vis., 2020, pp. 660-676

    2020

  55. [63]

    Distance map loss penalty term for semantic segmentation,

    F. Caliva, C. Iriondo, A. M. Martinez et al., “Distance map loss penalty term for semantic segmentation,” Unpublished paper, 2019. [Online]. Available: https://arxiv.org/abs/1908.03679

    2019 arXiv

  56. [64]

    Boundary loss for highly unbalanced segmentation

    H. Kervadec, J. Bouchtiba, C. Desrosiers, E. Granger, J. Dolz and I. B. Ayed, "Boundary loss for highly unbalanced segmentation", in Proc. Int. Conf. Med. Imag. Deep Learn., 2019, pp. 285-296

    2019

  57. [65]

    A new spatial-oriented object detection framework for remote sensing images,

    D. Yu, and S. Ji, “A new spatial-oriented object detection framework for remote sensing images,” IEEE Trans. Geosci. Remote. Sens., vol. 60, pp. 1-16, 2021

    2021

  58. [66]

    UNet++: A nested u-net architecture for medical image segmentation

    Z. Zhou, M. M. R. Siddiquee, N. Tajbakhsh and J. Liang, "UNet++: A nested u-net architecture for medical image segmentation", in Proc. Int. Workshop Multimodal Learn. Clin. Decis. Support, vol. 11045, pp. 3-11, Jun. 2018

    2018

  59. [67]

    Instance normalization: The missing ingredient for fast stylization,

    D. Ulyanov, A. Vedaldi, and V. Lempitsky, “Instance normalization: The missing ingredient for fast stylization,” Unpublished paper, 2016. [Online]. Available: https://arxiv.org/abs/1607.08022

    2016 arXiv

  60. [68]

    Fully convolutional networks for multisource building extraction from an open aeri al and satellite imagery data set,

    S. Ji, S. Wei, and M. Lu, “Fully convolutional networks for multisource building extraction from an open aeri al and satellite imagery data set,” IEEE Trans. Geosci. Remote Sens., vol. 57, no. 1, pp. 574-586, 2018

    2018

  61. [69]

    A dataset of building instances of typical cities in China,

    K. Wu, D. Zheng, Y. Chen et al., “A dataset of building instances of typical cities in China,” Chinese Sci. Data, vol. 6, pp. 191-199, 2021

    2021

  62. [70]

    Microsoft coco captions: Data collection and evaluation server,

    X. Chen, H. Fang, T.-Y. Lin et al., “Microsoft coco captions: Data collection and evaluation server,” Unpublished paper, 2015. [Online]. Available: https://arxiv.org/abs/1504.00325. [56] Y. Lee, and J. Park, "Centermask: Real-time anchor-free instance segmentation," in Proc. I...

    2015 arXiv

  63. [71]

    Object detection and instance segmentation in remote sensing imagery based on precise mask R- CNN

    H. Su, S. Wei, M. Yan et al., "Object detection and instance segmentation in remote sensing imagery based on precise mask R- CNN." in Proc. IEEE Int. Geosci. Remote Sens. Symp. (IGARSS), 2019, pp. 1454-1457

    2019

  64. [72]

    DCTC: Fast and Accurate Contour-Based Instance Segmentation with DCT Encoding for High Resolution Remote Sensing Images,

    Z. Chen, T. Liu, X. Xu et al., “DCTC: Fast and Accurate Contour-Based Instance Segmentation with DCT Encoding for High Resolution Remote Sensing Images,” IEEE J. Sel. Top. Appl. Earth Observ. Remote Sens. , pp. 8697-8709, 2024

    2024

Pith tools

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