Pith. sign in

REVIEW 5 major objections 6 minor 41 references

OSDMamba: Enhancing Oil Spill Detection from Remote Sensing Images Using Selective State Space Model

T0 review · 5 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read This paper proposes OSDMamba, the first Mamba-based architecture for oil spill detection, and reports state-of-the-art segmentation accuracy on two public remote sensing datasets.

desk verdict A clean, competent first application of Mamba to oil spill segmentation, but the SOTA claim depends on baseline comparisons the paper doesn't show are controlled. read the letter →

arxiv 2506.18006 v1 pith:ETXV4PJO submitted 2025-06-22 cs.CV

classification cs.CV
keywords oilspilldetectionremotesensingsemanticsegmentationstatespacemodelsMambaclassimbalanceSARimagerymulti-scalefeaturefusion
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

The paper argues that convolutional networks fail on oil spill detection because their small receptive fields miss global context, causing small and minority-class spill regions to be missed or misclassified. To fix this, it introduces OSDMamba, a segmentation model built on Mamba's selective state space scanning, which expands the receptive field while preserving detail. The model pairs this encoder with an asymmetric decoder using ConvSSM and deep supervision to improve multi-scale feature fusion and sensitivity to rare classes. On the M4D SAR dataset the paper reports a mIoU of 70.25%, and on the MADOS Sentinel-2 dataset it reports an F1 of 71.2% and mIoU of 68.1%, beating all compared baselines. If correct, this shows that state space models are a practical alternative to CNNs and transformers for sparse-target remote sensing segmentation.

What carries the argument

The central object is the Vision State Space (VSS) Block, which uses 2D Selective Scan (SS2D) to unfold image patches in four directions, run each sequence through the selective state space model, and merge the results into a 2D feature map. This mechanism gives the model a global receptive field at linear-time cost, which the paper argues is what lets it detect small oil spills that convolution misses. The other load-bearing component is the asymmetric decoder: early decoding stages combine VSS Blocks with ConvSSM, a convolutional state space model whose kernel is restricted to pointwise convolution and which is mathematically equivalent to a block-diagonal SSM, while later stages use patch expansion with dual VSS Blocks; deep supervision at 1/4, 1/8, and 1/16 scales guides multi-scale representation learning.

What would settle it

Train U-Net, SegNext, and MariNeXt on M4D with the same AdamW settings, 100 epochs, hybrid loss, ImageNet initialization, and the same 110-image test split as OSDMamba; if any of these retrained baselines matches or exceeds OSDMamba's mIoU of 70.25%, the reported SOTA claim collapses.

Watch

Extended reading notes

Core claim

OSDMamba achieves a mIoU of 70.25% on the M4D dataset, outperforming all models in the comparison, with a 12.18% improvement in oil-spill-class IoU over the second-best model, U-Net. On the MADOS dataset it surpasses all competing methods, including the reproduced MariNeXt baseline, with an improvement of 0.6% in F1-score and 8.9% in mIoU. The paper attributes these gains to the VSS encoder's 2D selective scan, which aggregates sparse signals across the spatial domain and preserves boundaries of small spills, and to an asymmetric decoder whose ConvSSM and deep supervision strengthen multi-scale fusion and minority-class learning. It also reports fewer false positives than a U-Net baseline on MADOS and lower parameter and FLOP counts than a similar-scale Mix Transformer.

Load-bearing premise

The central claim assumes that all comparison models were trained and evaluated under the same protocol and tuning effort as OSDMamba; the paper details OSDMamba's training but not the baselines', and it uses a reproduced MariNeXt without describing the reproduction.

Editorial extensions

If this is right

  • Mamba-based encoders can outperform CNN and Transformer baselines on oil spill segmentation, offering a new direction for marine remote sensing.
  • The large oil-spill-class IoU improvement over U-Net (12.18%) suggests that selective scanning helps models learn underrepresented classes in imbalanced datasets.
  • Lower false-positive rates on MADOS indicate that the architecture reduces confusion between oil spills and look-alike natural phenomena.
  • With 110M parameters and about 270G FLOPs, the model offers a more efficient alternative to a similar-scale transformer-based segmenter, which may ease deployment on satellite data pipelines.
  • The architecture transfers design patterns from medical image segmentation (VSS blocks, deep supervision, asymmetric decoding) to a remote sensing task, suggesting these patterns generalize beyond their original domain.

Reading between the lines

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

  • Because the paper does not report training hyperparameters for the baselines or describe how MariNeXt was reproduced, the magnitude of the reported SOTA gains should be treated as provisional until a fair-protocol rerun confirms them.
  • The selective scan may be especially beneficial for thin, elongated oil slicks rather than blob-like spills; this could be tested by evaluating OSDMamba on per-shape subsets of the test images.
  • The same encoder-decoder recipe could transfer to other sparse-target remote sensing tasks, such as ship detection or marine debris segmentation, where class imbalance and small objects dominate.
  • A direct ablated comparison between SS2D scanning and an equivalent global attention module on the same decoder would clarify whether the gains come from the state space mechanism itself or from the decoder design.
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

5 major / 6 minor

Summary. The paper proposes OSDMamba, a Mamba/SSM-based encoder-decoder architecture for oil spill detection in remote sensing images. The encoder is built from Visual State Space (VSS) blocks following VMamba-Tiny, and the decoder combines ConvSSM, VSS blocks, patch expansion, and deep supervision. The method is evaluated on the M4D SAR dataset and the MADOS multispectral dataset, where it is reported to achieve state-of-the-art mIoU (70.25% on M4D, 68.1% on MADOS) and F1 (71.2% on MADOS). The paper also reports ablation studies and a false-positive analysis. The central claim is that OSDMamba outperforms all compared methods on both datasets, with improvements of 8.9% mIoU on MADOS and 11.8% IoU on the oil-spill class on M4D.

Significance. If the reported results hold under controlled comparison, OSDMamba would be a credible early application of state-space models to oil spill segmentation and would demonstrate the benefit of selective scanning for small-object and imbalanced-class problems. The paper gives credit for proposing a plausible architecture, including an asymmetric decoder with deep supervision, and for including ablations and a false-positive analysis. However, the empirical evidence is not airtight: the baseline training protocols are not specified, the key MariNeXt baseline is an undescribed reproduction, no error bars are reported, and no Mamba-based segmentation baselines are compared. These gaps make the magnitude of the claimed gains uncertain and limit the paper's immediate contribution as a reliable SOTA result.

major comments (5)
  1. [III-B and III-C] The comparison with state-of-the-art is not controlled. Section III-B specifies OSDMamba's training protocol (AdamW, learning rate 0.01, weight decay 0.0001, batch size 4, 100 epochs, hybrid loss) but gives no training protocol for the baselines in Tables I and II (U-Net, LinkNet, PSPNet, DeepLabv2/v3+, SAM-OIL, SegNext, MariNeXt). In particular, it is not stated whether the baselines used the same data splits, image sizes, augmentation, loss, or number of epochs. The 11.8% oil-spill IoU gain over U-Net and the 8.9% mIoU gain on MADOS are only evidence for the architecture if the baselines were given comparable training effort. Please provide the training configuration for each baseline, ideally using the same protocol, and report the corresponding hyperparameters.
  2. [III-C, Table II] The 'MariNeXt (reproduced)' row is load-bearing. OSDMamba's claimed 8.9% mIoU improvement on MADOS is essentially the gap to this reproduced number (68.1 vs 59.2), yet the reproduction is not described: no data split, number of epochs, loss function, optimizer, or image resolution, and no comparison to the original MariNeXt results reported in [40]. If the reproduction is not faithful, the reported SOTA margin is not meaningful. Please describe the reproduction protocol in detail and, if possible, use the original authors' evaluation setup or report both the original and reproduced numbers.
  3. [III-C, Table I] No error bars or multiple-seed variability are reported. The M4D mIoU margin over SAM-OIL is 70.25% vs 69.52%, a difference of 0.73 percentage points, which is within the typical run-to-run variance for semantic segmentation on this scale. Report mean and standard deviation over at least three independent runs for OSDMamba and the strongest baselines, and indicate whether the differences are statistically significant.
  4. [II-B and III-B] The proposed decoder and loss are not specified at the level needed for reproduction. Equations (3) and (4) describe a generic ConvSSM, but the manuscript does not state how ConvSSM is instantiated in OSDMamba (kernel sizes, number of channels, normalization, or how it is combined with the VSS blocks); the phrase 'Our variant, ConvS5' in Section II-B is undefined. In addition, the hybrid loss in Section III-B contains free parameters α_t and γ whose values are never reported. Please provide a complete layer-by-layer description of the decoder, the loss hyperparameters, and the exact scanning configuration.
  5. [I and III-C] The paper claims that OSDMamba is 'the first Mamba-based architecture specifically designed for oil spill detection' and that it outperforms all competing methods, but it does not compare with any Mamba-based segmentation baselines, such as VM-UNet [20], Swin-UMamba [19], or other recent SSM backbones. Without such comparisons, the improvement cannot be attributed to the specific OSDMamba design rather than to the general Mamba architecture. Please add at least one or two recent Mamba-based segmentation models as baselines.
minor comments (6)
  1. [Table I] Several entries have missing spaces between numbers (e.g., '51.6055.60', '52.5591.81', '96.47 65.5947.57 46.8594.76'); please format all table cells consistently.
  2. [Abstract] The abstract states 'improvements of 8.9% and 11.8% in OSD,' but this mixes two different metrics: 8.9% is the mIoU gain on MADOS, while 11.8% is the oil-spill-class IoU gain on M4D. Please state both metrics explicitly.
  3. [I] The introduction describes 'Mamba's sliding window mechanism for local self-attention,' which is inaccurate; Mamba uses a selective scan over token sequences, not self-attention. Please correct this description.
  4. [References] References [35] and [37] are the same paper (Chen et al., ECCV 2018); duplicate references should be removed.
  5. [III-A] The MADOS description mentions 15 thematic categories, but Table II reports only a single F1/mIoU; please clarify whether the metric is computed for the oil-spill class only and how the other categories are handled.
  6. [III-E] The qualitative analysis refers to '5th column of Fig. 4' but also says 'third columns'; the column references are confusing and should be rechecked.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: OSDMamba's results are test-set evaluations against external baselines, with no fitted-parameter-as-prediction or self-citation chain.

full rationale

The paper contains no derivation chain in which a claimed result is equivalent to an input by construction. Section III-C reports mIoU and F1 scores on held-out test sets of M4D and MADOS. These are empirical measurements of a trained model, not predictions derived from fitted parameters. No equation in the paper reduces a reported metric to a training objective or to the model's own hyperparameters. The architecture borrows VSS Blocks, ConvSSM, and decoder design from prior work (VMamba, Swin-UMamba, ConvSSM), but those citations are independent support for component choices, not load-bearing self-citations that force the outcome. No self-citations by the authors appear in the reference list. The only potentially load-bearing comparison is the reproduced MariNeXt baseline in Table II, but the absence of training details for that baseline is an evaluation-validity concern, not circularity: the baseline number is not defined in terms of OSDMamba's output and is not equivalent to the claimed improvement. Ablation experiments compare internal variants on the same test set, which is standard practice and does not constitute circular reasoning. Therefore, no circular step can be exhibited, and the appropriate score is 0.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new theoretical entities or fitted constants beyond standard deep learning hyperparameters. The model's weights are trained on the evaluation datasets, which is normal for empirical ML. The main unstated premises are about dataset quality and fair baseline comparison.

free parameters (5)
  • Focal loss alpha_t class weight = not reported
    Used in hybrid loss (Section III-B) to balance classes; value chosen by hand, not stated.
  • Focal loss gamma = not reported
    Modulation factor in hybrid loss (Section III-B).
  • Learning rate = 0.01
    AdamW optimizer setting (Section III-B).
  • Weight decay = 0.0001
    AdamW setting (Section III-B).
  • VSS block configuration = {2, 2, 9, 2}
    Encoder depth configuration taken from VMamba-Tiny (Section II-A2).
assumptions (4)
  • domain assumption ImageNet pretrained weights transfer to SAR and Sentinel-2 imagery
    The encoder is initialized with ImageNet weights (Section III-B); the reported performance depends on this transfer being effective for remote sensing modalities.
  • domain assumption The M4D and MADOS ground truth labels are correct
    All metrics are computed against these labels (Section III-A); any labeling errors directly affect the reported improvements.
  • domain assumption The reproduced MariNeXt baseline matches the original method
    MADOS comparison (Table II) uses 'MariNeXt (reproduced)' without describing the reproduction protocol; if the reproduction is unfaithful, the 8.9% mIoU gain is not a valid comparison.
  • standard math The VSS and ConvSSM implementations are correct as published
    The method builds directly on VSS blocks from VMamba [27] and ConvSSM from [12]; errors in those implementations would propagate.

how reviews work

0 comments
Cite this review

Pith. "Pith review of OSDMamba: Enhancing Oil Spill Detection from Remote Sensing Images Using Selective State Space Model." pith.science (2026). https://pith.science/paper/ETXV4PJO

@misc{pith2026250618006,
  author       = {Pith},
  title        = {Pith review of: OSDMamba: Enhancing Oil Spill Detection from Remote Sensing Images Using Selective State Space Model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ETXV4PJO}},
  note         = {Machine review of arXiv:2506.18006}
}
read the original abstract

Semantic segmentation is commonly used for Oil Spill Detection (OSD) in remote sensing images. However, the limited availability of labelled oil spill samples and class imbalance present significant challenges that can reduce detection accuracy. Furthermore, most existing methods, which rely on convolutional neural networks (CNNs), struggle to detect small oil spill areas due to their limited receptive fields and inability to effectively capture global contextual information. This study explores the potential of State-Space Models (SSMs), particularly Mamba, to overcome these limitations, building on their recent success in vision applications. We propose OSDMamba, the first Mamba-based architecture specifically designed for oil spill detection. OSDMamba leverages Mamba's selective scanning mechanism to effectively expand the model's receptive field while preserving critical details. Moreover, we designed an asymmetric decoder incorporating ConvSSM and deep supervision to strengthen multi-scale feature fusion, thereby enhancing the model's sensitivity to minority class samples. Experimental results show that the proposed OSDMamba achieves state-of-the-art performance, yielding improvements of 8.9% and 11.8% in OSD across two publicly available datasets.

Figures

Figures reproduced from arXiv: 2506.18006 by the authors.

Figure 1
Figure 1. The distribution of the semantic classes in the Oil spill detection [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Architectural overview of the proposed method. OSDMamba uses a pre-trained, modified VMamba as the encoder. The decoder consists of two parts: [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Comparison of receptive fields between convolution and 2D selective [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Qualitative analysis of ablation experiments [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

41 extracted references · 35 canonical work pages

  1. [40]

    Detecting Marine pollutants and Sea Surface features with Deep learning in Sentinel-2 imagery[J]

    Kikaki K, Kakogeorgiou I, Hoteit I, et al. Detecting Marine pollutants and Sea Surface features with Deep learning in Sentinel-2 imagery[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2024, 210: 39- 54

  2. [20]

    Vm-unet: Vision mamba unet for medical image segmentation,

    J. Ruan and S. Xiang, “Vm-unet: Vision mamba unet for medical image segmentation,”arXiv preprint arXiv:2402.02491, 2024

  3. [1]

    Oil spill identification,

    A. P. Bentz, “Oil spill identification,”Analytical Chemistry, vol. 48, no. 6, pp. 454A-472A, 1976

  4. [2]

    J. W. Doerffer,Oil spill response in the marine environment, Elsevier, 2013

  5. [3]

    Sensors, features, and machine learning for oil spill detection and monitoring: A review,

    R. Al-Ruzouq, M. B. A. Gibril, A. Shanableh, et al., “Sensors, features, and machine learning for oil spill detection and monitoring: A review,” Remote Sensing, vol. 12, no. 20, pp. 3338, MDPI, 2020

  6. [4]

    Oil spill detection using marine SAR images,

    B. Fiscella, A. Giancaspro, F. Nirchio, P. Pavese, and P. Trivero, “Oil spill detection using marine SAR images,”International Journal of Remote Sensing, vol. 21, no. 18, pp. 3561-3566, Taylor & Francis, 2000

  7. [5]

    Oil spill detection using synthetic aperture radar images and feature selection in shape space,

    Y . Guo and H. Z. Zhang, “Oil spill detection using synthetic aperture radar images and feature selection in shape space,”International Journal of Applied Earth Observation and Geoinformation, vol. 30, pp. 146-157, Elsevier, 2014

  8. [6]

    Advances in remote sensing technology, machine learning and deep learning for marine oil spill detection, prediction and vulnerability assessment,

    S. Temitope Yekeen and A.-L. Balogun, “Advances in remote sensing technology, machine learning and deep learning for marine oil spill detection, prediction and vulnerability assessment,”Remote Sensing, vol. 12, no. 20, pp. 3416, MDPI, 2020

Show all 41 references
  1. [7]

    U-net: Convolutional networks for biomedical image segmentation,

    O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image segmentation,” inMedical image computing and computer-assisted intervention–MICCAI 2015: 18th international confer- ence, Munich, Germany, October 5-9, 2015, proceedings, part III 18, ...

  2. [8]

    Oil spills detection from SAR Earth observations based on a hybrid CNN transformer networks,

    S. Dehghani-Dehcheshmeh, M. Akhoondzadeh, and S. Homayouni, “Oil spills detection from SAR Earth observations based on a hybrid CNN transformer networks,”Marine Pollution Bulletin, vol. 190, pp. 114834, Elsevier, 2023

  3. [9]

    DS-UNet: Dual-Stream U-Net for Oil Spill Detection of SAR Image,

    C. Li, M. Wang, X. Yang, and D. Chu, “DS-UNet: Dual-Stream U-Net for Oil Spill Detection of SAR Image,”IEEE Geoscience and Remote Sensing Letters, vol. 20, pp. 1-5, IEEE, 2023

  4. [10]

    Oil spill detection by satellite remote sensing,

    C. Brekke and A.H.S. Solberg, “Oil spill detection by satellite remote sensing,”Remote sensing of environment, vol. 95, no. 1, pp. 1-13, Elsevier, 2005

  5. [11]

    Oil spill detection in Radarsat and Envisat SAR images,

    A.H.S. Solberg, C. Brekke, and P.O. Husoy, “Oil spill detection in Radarsat and Envisat SAR images,”IEEE Transactions on Geoscience and Remote Sensing, vol. 45, no. 3, pp. 746-755, IEEE, 2007

  6. [12]

    Con- volutional state space models for long-range spatiotemporal modeling,

    J. Smith, S. De Mello, J. Kautz, S. Linderman, and W. Byeon, “Con- volutional state space models for long-range spatiotemporal modeling,” Advances in Neural Information Processing Systems, vol. 36, 2024

  7. [13]

    Mamba: Linear-time sequence modeling with selective state spaces,

    A. Gu and T. Dao, “Mamba: Linear-time sequence modeling with selective state spaces,”arXiv preprint arXiv:2312.00752, 2023

  8. [14]

    Mixed Transformer U-Net for medical image segmentation,

    H. Wang, S. Xie, L. Lin, Y . Iwamoto, X.-H. Han, Y .-W. Chen, and R. Tong, “Mixed Transformer U-Net for medical image segmentation,”arXiv preprint arXiv:2111.04734, 2021

  9. [17]

    Oil spill identification from satellite images using deep neural networks,

    M. Krestenitis, G. Orfanidis, K. Ioannidis, K. Avgerinakis, S. Vrochidis, and I. Kompatsiaris, “Oil spill identification from satellite images using deep neural networks,”Remote Sensing, vol. 11, no. 15, pp. 1762, MDPI, 2019

  10. [18]

    nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation,

    F. Isensee, P.F. Jaeger, S.A.A. Kohl, J. Petersen, and K.H. Maier-Hein, “nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation,”Nature methods, vol. 18, no. 2, pp. 203-211, Nature Publishing Group, 2021

  11. [21]

    Swin transformer: Hierarchical vision transformer using shifted windows,

    Z. Liu, Y . Lin, Y . Cao, H. Hu, Y . Wei, Z. Zhang, S. Lin, and B. Guo, “Swin transformer: Hierarchical vision transformer using shifted windows,” inProceedings of the IEEE/CVF international conference on computer vision, pp. 10012-10022, 2021

  12. [22]

    Image transformer,

    N. Parmar, A. Vaswani, J. Uszkoreit, L. Kaiser, N. Shazeer, A. Ku, and D. Tran, “Image transformer,” inInternational conference on machine learning, pp. 4055-4064, PMLR, 2018

  13. [23]

    Sensors, features, and machine learning for oil spill detection and monitoring: A review,

    R. Al-Ruzouq, M.B.A. Gibril, A. Shanableh, A. Kais, O. Hamed, S. Al-Mansoori, and M.A. Khalil, “Sensors, features, and machine learning for oil spill detection and monitoring: A review,”Remote Sensing, vol. 12, no. 20, pp. 3338, MDPI, 2020

  14. [24]

    Focal loss for dense object detection,

    T.-Y . Lin, P. Goyal, R. Girshick, K. He, and P. Doll ´ar, “Focal loss for dense object detection,” inProceedings of the IEEE International Conference on Computer Vision, 2017, pp. 2980–2988

  15. [25]

    Deep learning based on Transformer architecture for power system short-term voltage stability assessment with class imbalance,

    Y . Li, J. Cao, Y . Xu, L. Zhu, and D.Z. Dong, “Deep learning based on Transformer architecture for power system short-term voltage stability assessment with class imbalance,”Renewable and Sustainable Energy Reviews, vol. 189, pp. 113913, Elsevier, 2024

  16. [26]

    Power transformer fault diagnosis considering data imbalance and data set fusion,

    Y . Zhang, H.C. Chen, Y . Du, M. Chen, J. Liang, J. Li, X. Fan, and X. Yao, “Power transformer fault diagnosis considering data imbalance and data set fusion,”High V oltage, vol. 6, no. 3, pp. 543-554, Wiley Online Library, 2021

  17. [27]

    Vision mamba: Efficient visual representation learning with bidirectional state space model,

    L. Zhu, B. Liao, Q. Zhang, X. Wang, W. Liu, and X. Wang, “Vision mamba: Efficient visual representation learning with bidirectional state space model,”arXiv preprint, arXiv:2401.09417, 2024

  18. [28]

    Offset curves loss for imbalanced problem in medical segmentation,

    N. Le, T. Le, K. Yamazaki, T. Bui, K. Luu, and M. Savides, “Offset curves loss for imbalanced problem in medical segmentation,” in2020 25th International Conference on Pattern Recognition (ICPR), pp. 9189- 9195, IEEE, 2021

  19. [29]

    Com- positional Oil Spill Detection Based on Object Detector and Adapted Segment Anything Model from SAR Images,

    W. Wu, M. Sing Wong, X. Yu, G. Shi, C.Y .T. Kwok, and K. Zou, “Com- positional Oil Spill Detection Based on Object Detector and Adapted Segment Anything Model from SAR Images,”IEEE Geoscience and Remote Sensing Letters, IEEE, 2024

  20. [30]

    Oil Spill Segmentation using Deep Encoder-Decoder models,

    A. R. Satyanarayana and M. A. Dhali, “Oil Spill Segmentation using Deep Encoder-Decoder models,”arXiv preprint, arXiv:2305.01386, 2023

  21. [31]

    Swin-umamba: Mamba-based unet with imagenet-based pretraining,

    J. Liu, H. Yang, H. Y . Zhou, et al., “Swin-umamba: Mamba-based unet with imagenet-based pretraining,”arXiv preprint, arXiv:2402.03302, 2024

  22. [32]

    Linknet: Exploiting encoder rep- resentations for efficient semantic segmentation,

    A. Chaurasia and E. Culurciello, “Linknet: Exploiting encoder rep- resentations for efficient semantic segmentation,” in2017 IEEE Visual Communications and Image Processing (VCIP), IEEE, pp. 1-4, 2017

  23. [33]

    Pyramid scene parsing network,

    H. Zhao, J. Shi, X. Qi, et al., “Pyramid scene parsing network,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2881-2890, 2017

  24. [34]

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

    L. C. Chen, G. Papandreou, I. Kokkinos, et al., “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

  25. [35]

    Encoder-decoder with atrous separable convolution for semantic image segmentation,

    L. C. Chen, Y . Zhu, G. Papandreou, et al., “Encoder-decoder with atrous separable convolution for semantic image segmentation,” inProceedings of the European Conference on Computer Vision (ECCV), pp. 801-818, 2018

  26. [36]

    Oil spill modeling: A critical review on current trends, perspectives, and challenges,

    P. Keramea, K. Spanoudaki, G. Zodiatis, et al., “Oil spill modeling: A critical review on current trends, perspectives, and challenges,”Journal of Marine Science and Engineering, vol. 9, no. 2, p. 181, 2021

  27. [37]

    C., Zhu, Y ., Papandreou, G., Schroff, F., & Adam, H

    Chen, L. C., Zhu, Y ., Papandreou, G., Schroff, F., & Adam, H. (2018). Encoder-decoder with atrous separable convolution for semantic image segmentation. In Proceedings of the European conference on Computer Vision (ECCV) (pp. 801-818)

  28. [38]

    J., Li, K., & Fei-Fei, L

    Deng, J., Dong, W., Socher, R., Li, L. J., Li, K., & Fei-Fei, L. (2009, June). Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition (pp. 248-255). Ieee

  29. [39]

    Loshchilov, I. (2017). Decoupled weight decay regularization. arXiv preprint arXiv:1711.05101

  30. [41]

    MARIDA: A benchmark for Marine Debris detection from Sentinel-2 remote sensing data[J]

    Kikaki K, Kakogeorgiou I, Mikeli P, et al. MARIDA: A benchmark for Marine Debris detection from Sentinel-2 remote sensing data[J]. PloS one, 2022, 17(1): e0262247

  31. [42]

    Segnext: Rethinking convolutional attention design for semantic segmentation[J]

    Guo M H, Lu C Z, Hou Q, et al. Segnext: Rethinking convolutional attention design for semantic segmentation[J]. Advances in Neural Infor- mation Processing Systems, 2022, 35: 1140-1156

  32. [43]

    Oil spills detection from SAR Earth observations based on a hybrid CNN trans- former networks[J]

    Dehghani-Dehcheshmeh S, Akhoondzadeh M, Homayouni S. Oil spills detection from SAR Earth observations based on a hybrid CNN trans- former networks[J]. Marine Pollution Bulletin, 2023, 190: 114834

  33. [44]

    Detection of Marine Oil Spill from PlanetScope Images Using CNN and Transformer Models[J]

    Kang J, Yang C, Yi J, et al. Detection of Marine Oil Spill from PlanetScope Images Using CNN and Transformer Models[J]. Journal of Marine Science and Engineering, 2024, 12(11): 2095

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.