Pith. sign in

REVIEW 3 major objections 4 minor 53 references

Label-Decoupled Style Augmentation for Domain Generalization in Multi-Label Remote Sensing Scene Classification

T0 review · 3 major / 4 minor · reviewed 2026-08-02 · deepseek-v4-flash

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

Pith's one-line read Attention-based per-label style mixing improves multi-label remote-sensing domain generalization (71.5% mAP; +5.0 over ERM, +1.3 over global-statistics baseline).

desk verdict First multi-label RS DG benchmark and a plausible per-label extension of MixStyle-family augmenters, but the headline edge over the strongest global baseline is within noise and partly confounded by per-variant tuning; needs significance tests and code before I'd trust the +5 mAP claim. read the letter →

arxiv 2607.12704 v3 pith:EJAN24SK submitted 2026-07-14 cs.CV cs.LG

Alaa Almouradi , Erchan Aptoula This is my paper · ORCID
classification cs.CVcs.LG
keywords multi-labelaugmentationgeneralizationstyleattentiondomainlabelsper-label
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

Multi-label aerial images contain several things at once: a coastal tile may show ships, water, and pavement together. Style augmentation for domain generalization usually mixes the global color/texture statistics of two training images, but in a multi-label tile that lets one object's appearance repaint another. The authors replace global statistics with per-label statistics. Attention maps, from a small learned module or from cached Grad-CAM, locate each label, and each label's statistics are mixed only with a cross-domain image that also contains that label, under an independent random mixing strength. Training changes; inference does not. They build a benchmark from UCM, AID, and DFC15 over six shared labels and test six variants on three leave-one-domain-out splits with five seeds. The best variant, LD-EFDMix-GC, reaches 71.5% mean average precision: 5.0 points over plain training and 1.3 points over the best global-statistics method, with a 7.7-point gain on the hardest transfer. Ablations show spatial attention and refreshed localization maps matter most. The caveats are that gains are modest relative to seed noise, no significance tests are reported, some hyperparameters were manually adjusted after proxy tuning, and code is promised only upon acceptance. The core idea is simple and cheap: if attention localizes classes well, separating style noise by label should give more realistic augmentation.
Extended reading notes

Core claim

Central claim: a label-decoupled style augmentation framework, using per-label attention, cross-domain label-matched statistics mixing, and independent per-label coefficients, improves multi-label remote-sensing domain generalization. Empirically: 'Averaged over three splits and five seeds, the best variant attains 71.5% mean average precision, exceeding empirical risk minimization by 5.0 points and the strongest global-statistics baseline by 1.3 points' (Abstract; Table 7). If correct, label decoupling is a generic, inference-free upgrade for MixStyle/EFDMix/CSU on multi-label scenes.

Load-bearing premise

The framework's engine is per-label attention; if attention is inaccurate, the per-label statistics are contaminated by other labels and the method degrades toward global mixing. The paper states this in Section 5.6: 'the method assumes that image-level labels suffice to localize label-specific style; in scenes where a class occupies very few pixels, the attention, learned or cached, may be too coarse.' This assumption is load-bearing in Eq. (3) (attention-weighted recomposition) and Eq. (4)/(6) (attention sources), and is structurally different from the claim that the method improves mAP. The ablation 'w/o attention (uniform map)' costs 2.6 mAP on Split 1, confirming the dependency.

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 / 4 minor

Summary. The paper proposes a label-decoupled style augmentation framework for multi-label remote sensing domain generalization. It decomposes feature statistics per label via attention (learned LLAM or cached Grad-CAM), mixes statistics only between cross-domain partners sharing the label with independent per-label coefficients, and recomposes features through attention-weighted normalization. Three operators (MixStyle, EFDMix, CSU) are instantiated with two attention sources, yielding six variants. The framework is evaluated on a leave-one-domain-out benchmark built from UCM, AID, and DFC15 with six shared labels. The best variant (LD-EFDMix-GC) achieves 71.5% mAP averaged over three splits and five seeds, +5.0 over ERM and +1.3 over the strongest global-statistics baseline (CSU). Ablations indicate that spatial attention and refreshed localization maps are the most influential components, and the method adds at most 0.35% parameters with no inference change.

Significance. If the empirical claim holds, the contribution is a generic, inference-free upgrade to feature-statistics augmentation for multi-label imagery, filling a real gap in multi-label remote sensing DG. The paper is strong in breadth: three splits, five seeds, eleven methods, component ablations, hyperparameter sensitivity, complexity analysis, and t-SNE visualization. The ablation isolating attention and map refresh is informative, and the limitation section explicitly acknowledges the dependency on attention quality and the small benchmark. However, the evaluation protocol has weaknesses that directly affect the central claim: per-method and per-split hyperparameter search with manual adjustments, no significance tests, and headline margins that are small relative to seed-to-seed variance. These issues currently prevent the paper from establishing that label decoupling itself, rather than tuning or selection, drives the reported gains.

major comments (3)
  1. [§4.5, Table 3] The hyperparameter-selection protocol cannot support the claim that label decoupling, rather than tuning, drives the gains. LOSO validation is constructed with only two sources by pairing the remaining source with an augmented version of itself as a pseudo-domain; 20 Optuna trials per method are run, and Section 4.5 states that for some LD variants the LOSO-selected values were 'further adjusted to push for more aggressive operations.' The global baselines' selected hyperparameters and search budgets are not reported. With per-split settings for LD-EFDMix-GC that vary strongly (e.g., α=0.45 vs 5.0, ρ=0.34 vs 0.60 in Table 3), the method is effectively picking favorable configurations per split. Please run the same search procedure for all baselines, or report results with fixed hyperparameters across splits, or use nested validation.
  2. [§5.1, Tables 4–7] The headline margin over the strongest global baseline is within plausible noise. The +1.3 mAP average in Table 7 aggregates three splits where LD-EFDMix-GC is best on Split 1 (79.5±2.4 vs CSU 76.1±1.5, Table 4), but worse than CSU on Split 2 (77.9±2.2 vs 78.2±1.0, Table 5), and within one standard deviation on Split 3 (57.3±5.1 vs 56.3±2.6, Table 6). No significance tests are provided; with five seeds and per-seed standard deviations of 2–5 mAP, a 1.3-point average difference is not statistically supported. Report paired per-seed differences across the common split/seed structure and provide a significance test (e.g., paired t-test or Wilcoxon signed-rank), both per split and pooled.
  3. [§4.1, §5.6] The benchmark is very small relative to the strength of the 'generic, inexpensive upgrade path' claim. Three domains yield only three leave-one-out splits with two source domains each; per-domain idiosyncrasies are likely to dominate. The paper acknowledges the limited scope in §5.6 ('six shared labels and three domains'), but the abstract's claim of a generic upgrade is not yet warranted. A domain-level analysis (e.g., bootstrapping over domains, or reporting per-domain uncertainty via leave-two-out variants) would help, and the abstract/conclusion should be tempered until such evidence is available.
minor comments (4)
  1. [§1, §2.2] The paper states 'no prior method or multi-source benchmark targets it' (multi-label RS DG), but §2.2 cites CLIPood and Mixup-CLIPood as multi-label DG methods. The 'first' claim should be qualified as 'first feature-statistics method for multi-label RS DG' or 'first multi-source multi-label benchmark for RS.'
  2. [Table 9] The column header '∆ MACs (G)' is confusing: the values shown are the same (4.087) for all methods and appear to be total MACs, not a delta. Please correct the label or report the actual delta.
  3. [§5.3] The hyperparameter sensitivity study uses single training runs at one seed, as acknowledged in §5.6. This is acceptable for coarse trends, but the text 'the method appears robust to its hyperparameters' is stronger than the evidence supports; consider adding a confidence interval or multi-seed points for at least the most influential parameter.
  4. [§4.4] The 'domain-balanced label-matched sampler' is mentioned as an important component, but its construction and interaction with the style bank are described only briefly. A precise algorithm or pseudo-code would improve reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's gains are held-out empirical results, not quantities defined from their own fitted inputs.

full rationale

The central claim is an empirical one: label-decoupled style augmentation improves multi-label remote-sensing domain generalization on held-out target splits. The framework's equations (1)-(13) are constructive definitions of the augmentation procedure; they do not encode the target-domain mAP or any fitted constant that is later reported as a prediction. Per-label statistics are computed from source features and source labels, mixing partners are constrained to differ in domain, and target labels/statistics never enter training. Evaluation is leave-one-domain-out on held-out target test subsets, with five seeds per split; the headline +5.0 mAP over ERM is a measured comparison, not a quantity implied by the method's definitions. Hyperparameter selection uses a LOSO proxy, and the paper discloses that some LD values were manually adjusted toward aggressive operations; this is a protocol/tuning strength concern of the Gulrajani-Lopez-Paz type, not circularity, because the search objective is source validation mAP and the reported numbers are on unseen target splits. The only self-citations ([43], [45]) appear in related-work enumerations of existing RS-DG efforts and are not load-bearing for the proposed framework; there is no invocation of the authors' prior results to justify the method's premises. No uniqueness theorem is imported, no ansatz is smuggled in via citation, and no equation reduces to its own output by construction. The paper's own limitations section acknowledges that attention may be too coarse for small classes and that target/style overlap reduces gains, but acknowledging an assumption is not circularity. The derivation chain is self-contained as an empirical study; correctness and benchmark-validity critiques should be directed to the evaluation protocol, not to circular reasoning.

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

The method's adjustable numbers are training hyperparameters (alpha, gate probability, insertion stage, top-K ratio, warm-up, refresh), tuned per split via LOSO then manually adjusted for some variants; they are not physical constants or model parameters in the traditional sense. The axioms are modeling assumptions: label-to-region correspondence, label-matched partner validity, and the claim that attention-weighted affine recomposition preserves the feature manifold. No new physical entities are introduced; LLAM and the style bank are software components rather than postulated theoretical entities.

free parameters (6)
  • Beta alpha for LD-EFDMix-GC = S1=0.45, S2=5.0, S3=5.0
    Beta distribution shape for per-label mixing coefficients; tuned per split by LOSO/Optuna (Table 3).
  • Gate probability p = S1=0.20, S2=0.80, S3=0.80
    Stochastic firing probability of each LD module; hyperparameter selected by search (Table 3).
  • Insertion stages K = S1={2}, S2={1}, S3={1}
    Which backbone stages receive LD modules; the most influential hyperparameter in sensitivity runs (Figure 4a).
  • Top-K ratio rho = S1=0.34, S2=0.60, S3=0.60
    Fraction of attention-selected locations used in LD-EFDMix sort matching; tuned per split (Table 3).
  • Refresh interval R = S1=9, S2=13, S3=13
    Epochs between Grad-CAM bank rebuilds; selected by search (Table 3).
  • Warm-up W = S1=9, S2=5, S3=5
    Epochs in global mode before per-label pathway activates; tuned per split (Table 3).
assumptions (4)
  • domain assumption Image-level labels suffice to localize label-specific style; attention can recover per-label regions.
    Explicitly acknowledged in Section 5.6; all per-label statistics (Eq. 2) and mixing (Eqs. 8-10) depend on this.
  • domain assumption Cross-domain samples sharing at least one label are valid style-transfer partners.
    Eq. (7) restricts pairing to y_i^T y_j >= 1; assumes shared-label semantics align across domains, which may fail when labels have different visual contexts.
  • ad hoc to paper Attention-weighted recomposition of per-label AdaIN branches stays on a valid feature manifold.
    Section 3.2 asserts the convex combination of affine branches preserves the manifold and is not an arbitrary distortion; this is asserted, not proven.
  • domain assumption Grad-CAM correctness-gating admits reliable localization maps.
    Section 3.3 stores maps only where the label is present and currently predicted correctly; rare or hard classes may lack entries, falling back to global statistics.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Label-Decoupled Style Augmentation for Domain Generalization in Multi-Label Remote Sensing Scene Classification." pith.science (2026). https://pith.science/paper/EJAN24SK

@misc{pith2026260712704,
  author       = {Pith},
  title        = {Pith review of: Label-Decoupled Style Augmentation for Domain Generalization in Multi-Label Remote Sensing Scene Classification},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EJAN24SK}},
  note         = {Machine review of arXiv:2607.12704}
}
read the original abstract

Multi-label classification assigns several co-occurring labels to each aerial scene, yet deployed models often encounter data distributions different from their training. Feature-statistics augmentation such as MixStyle, EFDMix, and correlated style uncertainty improves generalization at low cost but perturbs channel statistics globally, treating each image as a single style; one class can then contaminate the augmentation of another. Domain generalization is understudied for multi-label remote sensing; no prior method or multi-source benchmark targets it. A label-decoupled augmentation framework is therefore proposed, confining style perturbation to label-specific regions. Per-label attention, obtained from a learnable module or from gradient class-activation maps, yields per-label feature statistics; these statistics are mixed with cross-domain samples that share present labels, under independent per-label coefficients, and features are recomposed by attention-weighted normalization. Three operators combined with two attention sources produce six variants, evaluated on a leave-one-domain-out benchmark from multi-label UCM, AID, and DFC15 over six shared labels. Averaged over three splits and five seeds, the best variant attains 71.5% mean average precision, exceeding empirical risk minimization by 5.0 points and the strongest global-statistics baseline by 1.3 points, with the largest gain on the hardest transfer (up to 7.7 points). Ablations indicate that spatial attention and refreshed localization maps are most influential. The framework adds at most 0.35% parameters, leaves inference unchanged, and appears to offer a generic, inexpensive upgrade path for multi-label statistics-based domain generalization. Code is available upon acceptance at https://github.com/Alaa-Almouradi/Style-Augmentation-Upgrade.

Figures

Figures reproduced from arXiv: 2607.12704 by the authors.

Figure 1
Figure 1. Overview of the proposed LD style augmentation framework. Top: training pipeline [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. The three per-label augmentation operators explained in subsection 3.5. (a) LD-MixStyle [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Class samples from each domain in the benchmark, where the domain shift can be seen [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Hyperparameter sensitivity on Split 1 (target: UCM); each point is a single training run [PITH_FULL_IMAGE:figures/full_fig_p016_4.png]
Figure 5
Figure 5. Figure 5: t-SNE [53] embeddings of style statistics for an LD-EFDMix model on Split 1 (test parti￾tions of all domains). (a) Global per-image statistics (µ,σ), plotted once per (image, present label) pair for color comparability. (b) Per-label statistics (µℓ ,σℓ) from the propos…

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

53 extracted references · 1 linked inside Pith

  1. [1]

    Remote sensing image scene classification: Benchmark and state of the art.Proceedings of the IEEE, 105(10):1865–1883, 2017

    Gong Cheng, Junwei Han, and Xiaoqiang Lu. Remote sensing image scene classification: Benchmark and state of the art.Proceedings of the IEEE, 105(10):1865–1883, 2017

    2017

  2. [2]

    Multilabel remote sensing image retrieval using a semisupervised graph-theoretic method.IEEE Trans- actions on Geoscience and Remote Sensing, 56(2):1144–1158, 2018

    Bindita Chaudhuri, Beg¨ um Demir, Subhasis Chaudhuri, and Lorenzo Bruzzone. Multilabel remote sensing image retrieval using a semisupervised graph-theoretic method.IEEE Trans- actions on Geoscience and Remote Sensing, 56(2):1144–1158, 2018

    2018

  3. [3]

    Yuansheng Hua, Lichao Mou, and Xiao Xiang Zhu. Recurrently exploring class-wise atten- tion in a hybrid convolutional and bidirectional LSTM network for multi-label aerial image classification.ISPRS Journal of Photogrammetry and Remote Sensing, 149:188–199, 2019

    2019

  4. [4]

    A deep learning approach to UA V image multilabeling.IEEE Geoscience and Remote Sensing Letters, 14(5):694–698, 2017

    Abdallah Zeggada, Farid Melgani, and Yakoub Bazi. A deep learning approach to UA V image multilabeling.IEEE Geoscience and Remote Sensing Letters, 14(5):694–698, 2017

    2017

  5. [5]

    A deep multi-attention driven approach for multi-label remote sensing image classification.IEEE Access, 8:95934–95946, 2020

    Gencer Sumbul and Beg¨ um Demir. A deep multi-attention driven approach for multi-label remote sensing image classification.IEEE Access, 8:95934–95946, 2020

    2020

  6. [6]

    Relation network for multilabel aerial image classification.IEEE Transactions on Geoscience and Remote Sensing, 58(7):4558–4572, 2020

    Yuansheng Hua, Lichao Mou, and Xiao Xiang Zhu. Relation network for multilabel aerial image classification.IEEE Transactions on Geoscience and Remote Sensing, 58(7):4558–4572, 2020

    2020

  7. [7]

    Deep residual learning for im- age recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for im- age recognition. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 770–778, Las Vegas, NV, USA, June 2016

    2016

  8. [8]

    ImageNet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. ImageNet: A large-scale hierarchical image database. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 248–255, Miami, FL, USA, June 2009

    2009

Show all 53 references
  1. [9]

    Domain adaptation for the classification of remote sensing data: An overview of recent advances.IEEE Geoscience and Remote Sensing Magazine, 4(2):41–57, 2016

    Devis Tuia, Claudio Persello, and Lorenzo Bruzzone. Domain adaptation for the classification of remote sensing data: An overview of recent advances.IEEE Geoscience and Remote Sensing Magazine, 4(2):41–57, 2016

    2016

  2. [10]

    Domain-adversarial training of neural networks.Journal of Machine Learning Research, 17(59):1–35, 2016

    Yaroslav Ganin, Evgeniya Ustinova, Hana Ajakan, Pascal Germain, Hugo Larochelle, Fran¸ cois Laviolette, Mario Marchand, and Victor Lempitsky. Domain-adversarial training of neural networks.Journal of Machine Learning Research, 17(59):1–35, 2016

    2016

  3. [11]

    Generalizing from several related classifica- tion tasks to a new unlabeled sample

    Gilles Blanchard, Gyemin Lee, and Clayton Scott. Generalizing from several related classifica- tion tasks to a new unlabeled sample. InAdvances in Neural Information Processing Systems, pages 2178–2186, Granada, Spain, December 2011

    2011

  4. [12]

    Domain generalization: A survey.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(4):4396–4415, 2023

    Kaiyang Zhou, Ziwei Liu, Yu Qiao, Tao Xiang, and Chen Change Loy. Domain generalization: A survey.IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(4):4396–4415, 2023

    2023

  5. [13]

    Jindong Wang, Cuiling Lan, Chang Liu, Yidong Ouyang, Tao Qin, Wang Lu, Yiqiang Chen, Wenjun Zeng, and Philip S. Yu. Generalizing to unseen domains: A survey on domain gener- alization.IEEE Transactions on Knowledge and Data Engineering, 35(8):8052–8072, 2023. 19

    2023

  6. [14]

    Domain generalization via in- variant feature representation

    Krikamol Muandet, David Balduzzi, and Bernhard Sch¨ olkopf. Domain generalization via in- variant feature representation. InProceedings of the 30th International Conference on Machine Learning, pages 10–18, Atlanta, GA, USA, June 2013

    2013

  7. [15]

    Generalizing to unseen domains via adversarial data augmentation

    Riccardo Volpi, Hongseok Namkoong, Ozan Sener, John Duchi, Vittorio Murino, and Silvio Savarese. Generalizing to unseen domains via adversarial data augmentation. InAdvances in Neural Information Processing Systems, pages 5334–5344, Montr´ eal, Canada, December 2018

    2018

  8. [16]

    Carlucci, Antonio D’Innocente, Silvia Bucci, Barbara Caputo, and Tatiana Tommasi

    Fabio M. Carlucci, Antonio D’Innocente, Silvia Bucci, Barbara Caputo, and Tatiana Tommasi. Domain generalization by solving jigsaw puzzles. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 2229–2238, Long Beach, CA, USA, June 2019

    2019

  9. [17]

    A Fourier-based frame- work for domain generalization

    Qinwei Xu, Ruipeng Zhang, Ya Zhang, Yanfeng Wang, and Qi Tian. A Fourier-based frame- work for domain generalization. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 14383–14392, June 2021

    2021

  10. [18]

    Reducing domain gap by reducing style bias

    Hyeonseob Nam, HyunJae Lee, Jongchan Park, Wonjun Yoon, and Donggeun Yoo. Reducing domain gap by reducing style bias. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8690–8699, June 2021

    2021

  11. [19]

    SW AD: Domain generalization by seeking flat minima

    Junbum Cha, Sanghyuk Chun, Kyungjae Lee, Han-Cheol Cho, Seunghyun Park, Yunsung Lee, and Sungrae Park. SW AD: Domain generalization by seeking flat minima. InAdvances in Neural Information Processing Systems, pages 22405–22418, December 2021

    2021

  12. [20]

    Domain generalization with MixStyle

    Kaiyang Zhou, Yongxin Yang, Yu Qiao, and Tao Xiang. Domain generalization with MixStyle. InInternational Conference on Learning Representations, May 2021

    2021

  13. [21]

    Uncertainty modeling for out-of-distribution generalization

    Xiaotong Li, Yongxing Dai, Yixiao Ge, Jun Liu, Ying Shan, and Ling-Yu Duan. Uncertainty modeling for out-of-distribution generalization. InInternational Conference on Learning Rep- resentations, April 2022

    2022

  14. [22]

    Exact feature distribu- tion matching for arbitrary style transfer and domain generalization

    Yabin Zhang, Minghan Li, Ruihuang Li, Kui Jia, and Lei Zhang. Exact feature distribu- tion matching for arbitrary style transfer and domain generalization. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8035–8045, New Orleans, LA, US...

    2022

  15. [23]

    Domain generalization with correlated style uncertainty

    Zheyuan Zhang, Bin Wang, Debesh Jha, Beg¨ um Demir, and Ulas Bagci. Domain generalization with correlated style uncertainty. InProceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, pages 2000–2009, Waikoloa, HI, USA, January 2024

    2000

  16. [24]

    Toward generalizing to unseen domains in remote sensing: A survey.Neurocomputing, 2026

    Muxin Liao, Meiying Liao, Zhihong Lu, Yuting Sun, Yinglong Wang, and Yingqiong Peng. Toward generalizing to unseen domains in remote sensing: A survey.Neurocomputing, 2026

    2026

  17. [25]

    Spatial and structured SVM for multilabel image classification.IEEE Transactions on Geoscience and Remote Sensing, 56(10):5948–5960, 2018

    Satoru Koda, Abdallah Zeggada, Farid Melgani, and Ryuei Nishii. Spatial and structured SVM for multilabel image classification.IEEE Transactions on Geoscience and Remote Sensing, 56(10):5948–5960, 2018

    2018

  18. [26]

    Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, and Dhruv Batra

    Ramprasaath R. Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, and Dhruv Batra. Grad-CAM: Visual explanations from deep networks via gradient- based localization. InProceedings of the IEEE International Conference on Computer Vision, pages 618–626...

    2017

  19. [27]

    Bag-of-visual-words and spatial extensions for land-use classi- fication

    Yi Yang and Shawn Newsam. Bag-of-visual-words and spatial extensions for land-use classi- fication. InProceedings of the 18th SIGSPATIAL International Conference on Advances in Geographic Information Systems, pages 270–279, 2010

    2010

  20. [28]

    AID: A benchmark data set for performance evaluation of aerial scene classification.IEEE Transactions on Geoscience and Remote Sensing, 55(7):3965–3981, 2017

    Gui-Song Xia, Jingwen Hu, Fan Hu, Baoguang Shi, Xiang Bai, Yanfei Zhong, Liangpei Zhang, and Xiaoqiang Lu. AID: A benchmark data set for performance evaluation of aerial scene classification.IEEE Transactions on Geoscience and Remote Sensing, 55(7):3965–3981, 2017

    2017

  21. [29]

    [Online]

    2015 IEEE GRSS Data Fusion Contest. [Online]. Available:https://www.grss-ieee.org/ community/technical-committees/data-fusion, 2015

    2015

  22. [30]

    Multi-label classification: An overview.Interna- tional Journal of Data Warehousing and Mining, 3(3):1–13, 2007

    Grigorios Tsoumakas and Ioannis Katakis. Multi-label classification: An overview.Interna- tional Journal of Data Warehousing and Mining, 3(3):1–13, 2007

    2007

  23. [31]

    A review on multi-label learning algorithms.IEEE Transactions on Knowledge and Data Engineering, 26(8):1819–1837, 2014

    Min-Ling Zhang and Zhi-Hua Zhou. A review on multi-label learning algorithms.IEEE Transactions on Knowledge and Data Engineering, 26(8):1819–1837, 2014

    2014

  24. [32]

    CNN-RNN: A unified framework for multi-label image classification

    Jiang Wang, Yi Yang, Junhua Mao, Zhiheng Huang, Chang Huang, and Wei Xu. CNN-RNN: A unified framework for multi-label image classification. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 2285–2294, Las Vegas, NV, USA, June 2016

    2016

  25. [33]

    Learning spatial regularization with image-level supervisions for multi-label image classification

    Feng Zhu, Hongsheng Li, Wanli Ouyang, Nenghai Yu, and Xiaogang Wang. Learning spatial regularization with image-level supervisions for multi-label image classification. InProceed- ings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 5513–5522, Honolulu...

    2017

  26. [34]

    Multi-label image recognition with graph convolutional networks

    Zhao-Min Chen, Xiu-Shen Wei, Peng Wang, and Yanwen Guo. Multi-label image recognition with graph convolutional networks. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 5177–5186, Long Beach, CA, USA, June 2019

    2019

  27. [35]

    Deep learning for multilabel land cover scene categorization using data augmentation.IEEE Geoscience and Remote Sensing Letters, 16(7):1031–1035, 2019

    Radamanthys Stivaktakis, Grigorios Tsagkatakis, and Panagiotis Tsakalides. Deep learning for multilabel land cover scene categorization using data augmentation.IEEE Geoscience and Remote Sensing Letters, 16(7):1031–1035, 2019

    2019

  28. [36]

    Deep attention neural network for multi-label classification in unmanned aerial vehicle imagery

    Aaliyah Alshehri, Yakoub Bazi, Nassim Ammour, Haidar Almubarak, and Naif Alajlan. Deep attention neural network for multi-label classification in unmanned aerial vehicle imagery. IEEE Access, 7:119873–119880, 2019

    2019

  29. [37]

    Graph convolutional network for multi-label VHR remote sensing scene recognition.Neurocomputing, 357:36–46, 2019

    Nasir Khan, Ujjal Chaudhuri, Biplab Banerjee, and Subhasis Chaudhuri. Graph convolutional network for multi-label VHR remote sensing scene recognition.Neurocomputing, 357:36–46, 2019

    2019

  30. [38]

    CLIPood: Generalizing CLIP to out-of-distributions

    Yang Shu, Xingzhuo Guo, Jialong Wu, Ximei Wang, Jianmin Wang, and Mingsheng Long. CLIPood: Generalizing CLIP to out-of-distributions. InProceedings of the 40th International Conference on Machine Learning (ICML), volume 202 ofPMLR, pages 31716–31731, 2023

    2023

  31. [39]

    Robust domain generalization for multi-modal object recognition.arXiv preprint arXiv:2408.05831, 2024

    Yuxin Qiao, Keqin Li, Junhong Lin, Rong Wei, Chufeng Jiang, Yang Luo, and Haoyu Yang. Robust domain generalization for multi-modal object recognition.arXiv preprint arXiv:2408.05831, 2024. 21

    2024 arXiv

  32. [40]

    Single domain generalization for scene classification using style-oriented data augmentation

    Yimin Zhao, Guosheng Lin, Juepeng Zheng, Yang You, and Haohuan Fu. Single domain generalization for scene classification using style-oriented data augmentation. InIGARSS 2024 – 2024 IEEE International Geoscience and Remote Sensing Symposium, pages 8811– 8815, 2024

    2024

  33. [41]

    Frequency-based optimal style mix for domain generalization in semantic segmentation of remote sensing images.IEEE Transactions on Geoscience and Remote Sensing, 62:1–14, 2024

    Ryuhei Iizuka, Junshi Xia, and Naoto Yokoya. Frequency-based optimal style mix for domain generalization in semantic segmentation of remote sensing images.IEEE Transactions on Geoscience and Remote Sensing, 62:1–14, 2024

    2024

  34. [42]

    Style and content separation network for remote sensing image cross-scene generalization.ISPRS Journal of Photogrammetry and Remote Sensing, 201:1–11, 2023

    Sihan Zhu, Chen Wu, Bo Du, and Liangpei Zhang. Style and content separation network for remote sensing image cross-scene generalization.ISPRS Journal of Photogrammetry and Remote Sensing, 201:1–11, 2023

    2023

  35. [43]

    Band aware domain generalization for cross-country multispectral remote sensing scene classification

    Emre Durakli, Dilara Turan, Mamatha Thota, Petra Bosilj, and Erchan Aptoula. Band aware domain generalization for cross-country multispectral remote sensing scene classification. In 2024 14th Workshop on Hyperspectral Imaging and Signal Processing: Evolution in Remote Sensing ...

    2024

  36. [44]

    Ng, Bing Zhang, and Jocelyn Chanussot

    Zhu Han, Ce Zhang, Lianru Gao, Zhiqiang Zeng, Michael K. Ng, Bing Zhang, and Jocelyn Chanussot. Multisource collaborative domain generalization for cross-scene remote sensing image classification.IEEE Transactions on Geoscience and Remote Sensing, 62:1–15, 2024

    2024

  37. [45]

    Domain generalized object detection for remote sensing images

    Emre Durakli and Erchan Aptoula. Domain generalized object detection for remote sensing images. In2023 31st Signal Processing and Communications Applications Conference (SIU), pages 1–4, 2023

    2023

  38. [46]

    Language-aware domain generalization network for cross-scene hyperspectral image classification.IEEE Transactions on Geoscience and Remote Sensing, 61:1–12, 2023

    Yuxiang Zhang, Mengmeng Zhang, Wei Li, Shuai Wang, and Ran Tao. Language-aware domain generalization network for cross-scene hyperspectral image classification.IEEE Transactions on Geoscience and Remote Sensing, 61:1–12, 2023

    2023

  39. [47]

    Ziyang Gong, Zhixiang Wei, Di Wang, Xiaoxing Hu, Xianzheng Ma, Hongruixuan Chen, Yuru Jia, Yupeng Deng, Zhenming Ji, Xiangwei Zhu, Xue Yang, Naoto Yokoya, Jing Zhang, Bo Du, Junchi Yan, and Liangpei Zhang. Crossearth: Geospatial vision foundation model for domain generalizable...

    2026

  40. [48]

    Arbitrary style transfer in real-time with adaptive instance normalization

    Xun Huang and Serge Belongie. Arbitrary style transfer in real-time with adaptive instance normalization. InProceedings of the IEEE International Conference on Computer Vision, pages 1501–1510, Venice, Italy, October 2017

    2017

  41. [49]

    Learning deep features for discriminative localization

    Bolei Zhou, Aditya Khosla, Agata Lapedriza, Aude Oliva, and Antonio Torralba. Learning deep features for discriminative localization. InProceedings of the IEEE Conference on Com- puter Vision and Pattern Recognition, pages 2921–2929, Las Vegas, NV, USA, June 2016

    2016

  42. [50]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. InInternational Conference on Learning Representations, New Orleans, LA, USA, May 2019

    2019

  43. [51]

    In search of lost domain generalization

    Ishaan Gulrajani and David Lopez-Paz. In search of lost domain generalization. InInterna- tional Conference on Learning Representations, May 2021

    2021

  44. [52]

    Optuna: A next-generation hyperparameter optimization framework

    Takuya Akiba, Shotaro Sano, Toshihiko Yanase, Takeru Ohta, and Masanori Koyama. Optuna: A next-generation hyperparameter optimization framework. InProceedings of the 25th ACM 22 SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 2623– 2631, Anchorage...

    2019

  45. [53]

    Visualizing data using t-SNE.Journal of Machine Learning Research, 9:2579–2605, 2008

    Laurens van der Maaten and Geoffrey Hinton. Visualizing data using t-SNE.Journal of Machine Learning Research, 9:2579–2605, 2008. 23

    2008

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.