Pith. sign in

REVIEW 3 major objections 5 minor 40 references

A new four-class remote-sensing traffic dataset and a multi-scale network that keeps spatial and channel detail push traffic-object segmentation higher.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-11 22:49 UTC pith:LCJQINQQ

load-bearing objection Useful new four-class global traffic segmentation dataset with a competent multi-scale network; SOTA margin is thin and rests on a single split. the 3 major comments →

arxiv 2607.03945 v1 pith:LCJQINQQ submitted 2026-07-04 cs.CV

A Large-Scale Dataset and a New Method for RemoteSensing Traffic Object Segmentation

Zhigang Yang , Huiguang Yao , Linmao Tian , Qiang Li , Qi Wang This is my paper
classification cs.CV
keywords remote sensingtraffic object segmentationNWPU-Trafficmulti-scale feature fusioninstance-level annotationCSPNetspatial-channel preservation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Existing remote-sensing traffic datasets cover too few object types and scenes, so models cannot reliably measure real-world transportation capacity. The authors release NWPU-Traffic, 1,479 high-resolution images of cars, airplanes, ships and trains collected from 49 cities in seven countries after 2020, annotated with instance-level masks. On this benchmark they introduce CSPNet, which first interacts multi-scale features while explicitly preserving both spatial layout and channel semantics, then fuses local convolutional detail with global transformer context through a gated decoder. The resulting model records the highest mean IoU, overall accuracy and F1-score among seven strong baselines, and continues to segment dense, occluded or low-light traffic correctly on imagery never seen during training. The combination of a geographically broad, up-to-date dataset and an architecture that refuses to throw away either location or semantics is offered as a practical route to more trustworthy regional transport assessment.

Core claim

CSPNet, which keeps spatial position and channel semantics intact during multi-scale feature interaction and then adaptively blends local and global cues in the decoder, reaches state-of-the-art 73.50 % mIoU, 98.76 % overall accuracy and 83.36 % F1 on the NWPU-Traffic test set, outperforming DeepLabv3+, ABCNet, UNetFormer, GCBNet, ScaleFormer, CMLFormer and FSEL.

What carries the argument

Spatial-Channel Preserving Feature Interaction Network (CSPNet): pixel-shuffle / unfold operations that align multi-scale maps without collapsing position or channel identity, followed by a gated local-global decoder that weights convolutional detail against transformer context.

Load-bearing premise

That a single 7:1:2 geographic split of the 1,479 images already yields an unbiased picture of how the model will behave on cities or countries never represented in the collection.

What would settle it

Train the identical CSPNet on the same split, then measure mIoU on a completely held-out city or country outside the seven-country set; a large, systematic drop would falsify the claimed generalization.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper introduces NWPU-Traffic, a remote-sensing semantic-segmentation dataset of 1,479 images (GSD 0.12–0.5 m) containing 31,628 instance-level masks of four traffic categories (car, airplane, ship, train) collected from 49 cities in 7 countries, with most imagery dated 2020–2024. It also proposes CSPNet, a ResNet-34 encoder–decoder that inserts a Spatial-Preserved Feature Interaction Module (SPFIM), a Channel-Preserved Feature Interaction Module (CPFIM) that uses pixel-shuffle and frequency-domain weighting, and a Local-Global Feature Fusion Decoder (LGFFD) with gated CNN–Transformer fusion. On a single 7:1:2 geographic split the method reports 73.50 mIoU / 98.76 OA / 83.36 F1, edging CMLFormer by 0.55 mIoU, with ablations (Table 4) and qualitative results on dense, occluded and external untrained scenes offered as supporting evidence. Dataset and code are released.

Significance. A multi-category, multi-country traffic-object segmentation dataset with instance masks and recent imagery fills a genuine gap relative to detection-centric or single-class collections (HRSC2016, iSAID, SSDD, etc.). The architectural claim—that preserving spatial and channel information during multi-scale fusion improves small-object and dense-scene segmentation—is plausible and is backed by consistent ablations and external qualitative tests. Public release of data and code further raises the work’s utility for the remote-sensing community. If the reported ranking proves stable under multi-run and leave-one-city evaluation, the paper would become a useful benchmark and baseline for transportation-capacity analysis from satellite imagery.

major comments (3)
  1. Section 5.1 and Tables 2/3: the SOTA claim (CSPNet 73.50 mIoU vs. CMLFormer 72.95) rests on a single 7:1:2 geographic partition and one training run per method. No standard deviations, repeated seeds, or per-city / leave-one-country metrics are supplied. The absolute margin is only 0.55 mIoU overall and 0.09 IoU on the hardest class (car). Because residual city-level correlation can inflate all scores equally, the ranking is not yet shown to be stable; multi-run statistics or held-out-city numbers are required before the superiority claim can be accepted as load-bearing.
  2. Section 4.2 (CPFIM) and Section 4.3 (SPFIM): the central design premise is that pixel-shuffle / unshuffle and 3-D unfold operations “preserve” spatial and channel information that ordinary concat/sum lose. No quantitative diagnostic (e.g., boundary F-score, feature-map mutual information, or controlled ablation that isolates the preservation operators from the rest of the modules) is provided. Without such evidence the architectural novelty remains an untested assertion rather than a demonstrated mechanism.
  3. Table 1 and Section 3.5: NWPU-Traffic is presented as “large-scale” and the first comprehensive four-category traffic segmentation set. With 1,479 images it is smaller than several listed detection sets (xView, DOTA, HRRSD) and only modestly larger than iSAID’s traffic subset. The paper should either (a) justify the “large-scale” label by instance density / scene diversity metrics or (b) temper the claim so that the contribution is accurately scoped as a multi-category, multi-country instance-segmentation resource rather than a scale leap.
minor comments (5)
  1. Throughout the manuscript the dataset is alternately called “NWPU-Traffic” and “NPWU-Traffic” (e.g., Abstract vs. Sections 1–3). Standardize the acronym.
  2. Figure 5 is a composite of multiple overlapping architecture sketches; the final CSPNet topology is hard to parse. A single clean diagram with clear module boundaries would improve readability.
  3. Tables 2 and 3 are essentially duplicates (one with FLOPs/Params, one without). Merge them or drop the redundant table.
  4. Section 5.2: the loss is described as “binary cross entropy with Dice coefficient” without stating the weighting coefficient; please report the exact combination used.
  5. Several Chinese city names and figure captions contain minor English typos (“Shanhai”, “Canberr a”, “边缘” residual Chinese labels in early figure drafts). A language pass is needed.

Circularity Check

0 steps flagged

No circularity: new dataset + architecture evaluated empirically against external baselines on a held-out geographic split; no derivation reduces to its own inputs.

full rationale

The paper's central claims are (1) construction of the new NWPU-Traffic dataset (1 479 images, four traffic classes, 49 cities/7 countries, instance masks) and (2) empirical superiority of CSPNet (SPFIM + CPFIM + LGFFD) on that dataset's 7:1:2 geographic split (mIoU 73.50, OA 98.76, F1 83.36 vs. DeepLabv3+, ABCNet, UNetFormer, GCBNet, ScaleFormer, CMLFormer, FSEL). Neither claim is a mathematical derivation. There are no equations that define a quantity in terms of the quantity later 'predicted', no parameters fitted to a subset and then re-reported as predictions, no uniqueness theorems imported from the authors' prior work, and no ansatz smuggled via self-citation. Self-citations (e.g., prior road-extraction or multi-scale papers by the same group) appear only as background motivation and are not load-bearing for the reported numbers. Ablations (Table 4) and external qualitative tests (Figs. 10-12) are consistent with the architecture story but do not create circular dependence. The work is therefore self-contained against external benchmarks; circularity score is zero.

Axiom & Free-Parameter Ledger

3 free parameters · 3 axioms · 2 invented entities

The work is empirical computer vision. It inherits standard deep-learning practice (ResNet-34, AdamW, Dice loss, pixel-shuffle, FFT, Transformer blocks) and the usual remote-sensing assumption that polygon masks on optical imagery are a valid proxy for transportation capacity. No free physical constants or new particles are introduced; the only free parameters are ordinary training hyper-parameters.

free parameters (3)
  • base learning rate = 6e-4
    Set to 6e-4 with cosine annealing; chosen by the authors, not derived.
  • batch size / crop size = 4 / 512
    Batch 4, 512×512 random crops; engineering choices that affect final numbers.
  • loss weighting (BCE + Dice)
    Unweighted sum of binary cross-entropy and Dice; the relative scale is an implicit free parameter.
axioms (3)
  • domain assumption Standard supervised semantic-segmentation training with ResNet-34 ImageNet-pretrained encoder yields transferable multi-scale features for remote-sensing traffic objects.
    Invoked throughout Section 4 and 5; never proved, only empirically observed.
  • domain assumption Instance-level polygon masks on optical satellite imagery are a sufficient and unbiased proxy for evaluating regional transportation capacity.
    Stated in the introduction and abstract as the motivating use case.
  • ad hoc to paper Pixel-shuffle / unshuffle and 3-D unfold operations preserve the spatial and channel information needed for accurate multi-scale fusion.
    Core design claim of SPFIM and CPFIM (Sections 4.2–4.3); supported only by ablation, not by information-theoretic proof.
invented entities (2)
  • NWPU-Traffic dataset no independent evidence
    purpose: Provide the first multi-class, multi-country, instance-level traffic-object segmentation benchmark with post-2020 imagery.
    Newly collected and annotated; independent evidence will exist once the GitHub release is public and used by others.
  • CSPNet (SPFIM + CPFIM + LGFFD) no independent evidence
    purpose: Preserve spatial and channel cues while fusing multi-scale features and adaptively combining local/global context for traffic segmentation.
    Architectural construct introduced in Section 4; no external physical existence claimed.

pith-pipeline@v1.1.0-grok45 · 26836 in / 2788 out tokens · 20716 ms · 2026-07-11T22:49:04.256536+00:00 · methodology

0 comments
read the original abstract

Remote sensing imagery plays a crucial role in evaluating regional transportation capacity. However, existing segmentation datasets often lack diversity in object categories and scenes, limiting the ability of models to comprehensively evaluate trans portation capacity in real-world scenes. To alleviate this gap, we construct a large-scale and diverse dataset for transportation object segmentation, named as NWPU-Traffic. This dataset encompass four traffic object categories (car, airplane, ship, and train) and a wide range of scenes from 49 cities across 7 countries, with instance-level annotations to ensure precise segmentation of individual objects, which bridges critical shortcomings in resolution and scene diversity in existing datasets. Leveraging this dataset, we establish a benchmark with several popular segmentation networks. Furthermore, we propose a novel segmentation method that leverages spatial-channel preserving feature interaction and an adaptive feature decoder, enabling robust segmentation across varying scales and complex environments. Extensive experiments and ablation studies validate the effectiveness of our approach. The dataset and code are publicly available at https://github.com/CVer-Yang/NWPU-Traffic.

Figures

Figures reproduced from arXiv: 2607.03945 by Huiguang Yao, Linmao Tian, Qiang Li, Qi Wang, Zhigang Yang.

Figure 1
Figure 1. Figure 1: Transportation objects observed from satellite perspectives, such as airplanes, ships, cars, and [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: We select typical traffic scenes from seven countries globally for traffic target labeling, including [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: (A), (B), and (C) demonstrate the image count by category, target number distribution, and cumu [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: We show some representative scenes of the NWPU-Traffic and convert the segmentation masks of [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: The overall framework of the proposed CSPNet, and it is divided into four sections: the encoder, [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: Details of spatial-preserved feature interaction module. [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Visualization results of the model on the NWPU-Traffic dataset. [PITH_FULL_IMAGE:figures/full_fig_p019_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Visualization results of the model on the complex scenes. [PITH_FULL_IMAGE:figures/full_fig_p020_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Visualization results of the model on the Ports in the District of Columbia, Washington, USA. The [PITH_FULL_IMAGE:figures/full_fig_p022_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Visualization results of the model on the Ronald Reagan Washington National Airport, State of [PITH_FULL_IMAGE:figures/full_fig_p023_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Visualization results of the model on the Longcheng street, Taiyuan City, China. [PITH_FULL_IMAGE:figures/full_fig_p024_12.png] view at source ↗

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

40 extracted references · 1 canonical work pages

  1. [1]

    T. Yang, Q. Zhou, Q. Wang, Dia: Deriving linguistic information from auxil- iary languages for remote sensing image captioning, Pattern Recognition (2025) 112209

    2025

  2. [2]

    X. Chen, Q. Liu, H. H. Deng, T. Kuang, H. H.-Y . Lin, D. Xiao, J. Gateno, J. J. Xia, P.-T. Yap, Improving image segmentation with contextual and structural sim- ilarity, Pattern Recognition 152 (2024) 110489

    2024

  3. [3]

    Z. Yang, D. Zhou, Y . Yang, J. Zhang, Z. Chen, Road extraction from satellite imagery by road context and full-stage feature, IEEE Geoscience and Remote Sensing Letters 20 (2022) 1–5. 25

    2022

  4. [4]

    Zhang, K

    J. Zhang, K. Yang, A. Constantinescu, K. Peng, K. Müller, R. Stiefelhagen, Trans4Trans: Efficient transformer for transparent object and semantic scene seg- mentation in real-world navigation assistance, IEEE Transactions on Intelligent Transportation Systems 23 (10) (2022) 19173–19186

    2022

  5. [5]

    B. Ren, X. Yang, Y . Yu, J. Luo, Z. Deng, Pointobb-v2: Towards simpler, faster, and stronger single point supervised oriented object detection, arXiv preprint arXiv:2410.08210 (2024)

    Pith/arXiv arXiv 2024

  6. [6]

    Z. Liu, H. Wang, L. Weng, Y . Yang, Ship rotated bounding box space for ship ex- traction from high-resolution optical satellite images with complex backgrounds, IEEE Geoscience and Remote Sensing Letters 13 (8) (2016) 1074–1078

    2016

  7. [7]

    H. Zhu, X. Chen, W. Dai, K. Fu, Q. Ye, J. Jiao, Orientation robust object detection in aerial images using deep convolutional neural network, in: Proceedings of the IEEE International Conference on Image Processing, IEEE, 2015, pp. 3735– 3739

    2015

  8. [8]

    W. Lin, Z. Wu, J. Chen, J. Huang, L. Jin, Scale-aware modulation meet trans- former, in: Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 6015–6026

    2023

  9. [9]

    Zhang, X

    H. Zhang, X. Zuo, J. Jiang, C. Guo, J. Ma, MRFS: Mutually Reinforcing Im- age Fusion and Segmentation, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 26974–26983

    2024

  10. [10]

    J. Zhan, Y . Luo, C. Guo, Y . Wu, J. Meng, J. Liu, Yolopx: Anchor-free multi-task learning network for panoptic driving perception, Pattern Recognition 148 (2024) 110152

    2024

  11. [11]

    Cheng, P

    G. Cheng, P. Zhou, J. Han, Learning rotation-invariant convolutional neural net- works for object detection in VHR optical remote sensing images, IEEE Transac- tions on Geoscience and Remote Sensing 54 (12) (2016) 7405–7415. 26

    2016

  12. [12]

    Zhang, Y

    Y . Zhang, Y . Yuan, Y . Feng, X. Lu, Hierarchical and robust convolutional neural network for very high-resolution remote sensing object detection, IEEE Transac- tions on Geoscience and Remote Sensing 57 (8) (2019) 5535–5548

    2019

  13. [13]

    Z. Xiao, Q. Liu, G. Tang, X. Zhai, Elliptic fourier transformation-based his- tograms of oriented gradients for rotationally invariant object detection in remote- sensing images, International Journal of Remote Sensing 36 (2) (2015) 618–644

    2015

  14. [14]

    Zhang, X

    T. Zhang, X. Zhang, J. Li, X. Xu, B. Wang, X. Zhan, Y . Xu, X. Ke, T. Zeng, H. Su, et al., Sar ship detection dataset (ssdd): Official release and comprehensive data analysis, Remote Sensing 13 (18) (2021) 3690

    2021

  15. [15]

    Z. Zou, Z. Shi, Random access memories: A new paradigm for target detection in high resolution aerial remote sensing images, IEEE Transactions on Image Processing 27 (3) (2017) 1100–1111

    2017

  16. [16]

    Q. Li, Y . Yuan, Q. Wang, Multiscale factor joint learning for hyperspectral im- age super-resolution, IEEE Transactions on Geoscience and Remote Sensing 61 (2023) 1–10

    2023

  17. [17]

    Q. Li, M. Gong, Y . Yuan, Q. Wang, RGB-induced feature modulation network for hyperspectral image super-resolution, IEEE Transactions on Geoscience and Remote Sensing 61 (2023) 1–11

    2023

  18. [18]

    Zheng, Y

    Z. Zheng, Y . Zhong, J. Wang, A. Ma, Foreground-aware relation network for geospatial object segmentation in high spatial resolution remote sensing imagery, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recog- nition, 2020, pp. 4096–4105

    2020

  19. [19]

    Q. Zhu, X. Guo, Z. Li, D. Li, A review of multi-class change detection for satellite remote sensing imagery, Geo-spatial Information Science 27 (1) (2024) 1–15

    2024

  20. [20]

    Q. Li, M. Zhang, Z. Yang, Y . Yuan, Q. Wang, Edge-Guided Perceptual Network for Infrared Small Target Detection, IEEE Transactions on Geoscience and Re- mote Sensing (2024). 27

    2024

  21. [21]

    W. Yang, Y . Zhang, X. Liu, B. Gao, Scene adaptive building individual segmenta- tion based on large-scale airborne lidar point clouds, IEEE Transactions on Geo- science and Remote Sensing (2024)

    2024

  22. [22]

    Z. Yang, W. Zhang, Q. Li, W. Ni, J. Wu, Q. Wang, C 2 Net: Road Extraction via Context Perception and Cross Spatial-Scale Feature Interaction, IEEE Transac- tions on Geoscience and Remote Sensing (2024)

    2024

  23. [23]

    Z. Guo, H. Gan, CPP-Net: Embracing Multi-Scale Feature Fusion into Deep Unfolding CP-PPA Network for Compressive Sensing, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024, pp. 25086–25095

    2024

  24. [24]

    Huang, S

    H. Huang, S. Xie, L. Lin, Y . Iwamoto, X. Han, Y .-W. Chen, R. Tong, Scale- former: revisiting the transformer-based backbones from a scale-wise perspective for medical image segmentation, arXiv:2207.14552 (2022)

    Pith/arXiv arXiv 2022

  25. [25]

    H. Wu, P. Huang, M. Zhang, W. Tang, X. Yu, CMTFNet: Cnn and multiscale transformer fusion network for remote-sensing image semantic segmentation, IEEE Transactions on Geoscience and Remote Sensing 61 (2023) 1–12

    2023

  26. [26]

    Q. Zeng, J. Zhou, J. Tao, L. Chen, X. Niu, Y . Zhang, Multiscale global con- text network for semantic segmentation of high-resolution remote sensing images, IEEE Transactions on Geoscience and Remote Sensing 62 (2024) 1–13

    2024

  27. [27]

    L. Li, J. Yi, H. Fan, H. Lin, A lightweight semantic segmentation network based on self-attention mechanism and state space model for efficient urban scene segmentation, IEEE Transactions on Geoscience and Remote Sensing (2025). doi:10.1109/TGRS.2025.3562185

    work page doi:10.1109/tgrs.2025.3562185 2025

  28. [28]

    Heitz, D

    G. Heitz, D. Koller, Learning spatial context: Using stuff to find things, in: Pro- ceedings of the European Conference on Computer Vision, Springer, 2008, pp. 30–43

    2008

  29. [29]

    T. N. Mundhenk, G. Konjevod, W. A. Sakla, K. Boakye, A large contextual dataset for classification, detection and counting of cars with deep learning, in: 28 Proceedings of the European Conference on Computer Vision, Springer, 2016, pp. 785–800

    2016

  30. [30]

    Razakarivony, F

    S. Razakarivony, F. Jurie, Vehicle detection in aerial imagery: A small target de- tection benchmark, Journal of Visual Communication and Image Representation 34 (2016) 187–203

    2016

  31. [31]

    D. Lam, R. Kuzma, K. McGee, S. Dooley, M. Laielli, M. Klaric, Y . Bula- tov, B. McCord, xview: Objects in context in overhead imagery, arXiv preprint arXiv:1802.07856 (2018)

    Pith/arXiv arXiv 2018

  32. [32]

    G.-S. Xia, X. Bai, J. Ding, Z. Zhu, S. Belongie, J. Luo, M. Datcu, M. Pelillo, L. Zhang, DOTA: A large-scale dataset for object detection in aerial images, in: Proceedings of the IEEE Conference on Computer Vision and Pattern Recogni- tion, 2018, pp. 3974–3983

    2018

  33. [33]

    Airbus ship detection challenge dataset,https://www.kaggle.com/c/ airbus-ship-detection/, accessed: 2019-05-27 (2018)

    2019

  34. [34]

    Waqas Zamir, A

    S. Waqas Zamir, A. Arora, A. Gupta, S. Khan, G. Sun, F. Shahbaz Khan, F. Zhu, L. Shao, G.-S. Xia, X. Bai, isaid: A large-scale dataset for instance segmenta- tion in aerial images, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2019, pp. 28–37

    2019

  35. [35]

    L.-C. Chen, Y . Zhu, G. Papandreou, F. Schroff, H. Adam, Encoder-decoder with atrous separable convolution for semantic image segmentation, in: Proceedings of European Conference on Computer Vision, 2018, pp. 801–818

    2018

  36. [36]

    R. Li, S. Zheng, C. Zhang, C. Duan, L. Wang, 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 181 (2021) 84–98

    2021

  37. [37]

    L. Wang, R. Li, C. Zhang, S. Fang, C. Duan, X. Meng, P. M. Atkinson, UNet- Former: A unet-like transformer for efficient semantic segmentation of remote 29 sensing urban scene imagery, ISPRS Journal of Photogrammetry and Remote Sensing 190 (2022) 196–214

    2022

  38. [38]

    H. Wu, M. Zhang, P. Huang, W. Tang, CMLFormer: CNN and Multi-scale Local- context Transformer network for remote sensing images semantic segmentation, IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sens- ing (2024)

    2024

  39. [39]

    Q. Zhu, Y . Zhang, L. Wang, Y . Zhong, Q. Guan, X. Lu, L. Zhang, D. Li, A global context-aware and batch-independent network for road extraction from vhr satellite imagery, ISPRS Journal of Photogrammetry and Remote Sensing 175 (2021) 353–365

    2021

  40. [40]

    Y . Sun, C. Xu, J. Yang, H. Xuan, L. Luo, Frequency-spatial entanglement learning for camouflaged object detection, in: Proceedings of European Conference on Computer Vision, Springer, 2025, pp. 343–360. 30

    2025