Pith. sign in

REVIEW 3 major objections 4 minor 30 references

Investigating Location-Regularised Self-Supervised Feature Learning for Seafloor Visual Imagery

T0 review · 3 major / 4 minor · reviewed 2026-08-04 · 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 regularising self-supervised learning with location metadata—forming positive training pairs from physically nearby seafloor images rather than augmented views of one image—consistently improves downstream classificat

desk verdict Useful and honest empirical sweep, but the headline gains rest on single runs and the ViT claim outruns the evidence; worth refereeing with a demand for repeated seeds. read the letter →

arxiv 2509.06660 v1 pith:DSD3BKHT submitted 2025-09-08 cs.CV cs.RO

Cailei Liang , Adrian Bodenmann , Emma J Curtis , Samuel Simmons , Kazunori Nagano , Stan Brown , Adam Riese , Blair Thornton This is my paper · ORCID
classification cs.CVcs.RO
keywords seafloorimageryself-supervisedlearninglocationregularisationcontrastivevisiontransformersconvolutionalneuralnetworksautonomousunderwatervehiclesdownstreamclassification
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 asks whether the simple idea that nearby seafloor looks similar can make self-supervised learning more useful for robot-collected underwater imagery. It answers yes: replacing the standard SSL trick of pairing two augmented views of the same image with pairs of images taken from physically close locations improves downstream classifiers by about five F1 points for convolutional networks and six for vision transformers, averaged over three datasets. The gain is largest when latent representations are low-dimensional, and location-regularised CNNs can beat generic pretrained ones. The paper also finds that an off-the-shelf pretrained vision transformer generalises so well that dataset-specific fine-tuning, with or without location regularisation, only matches it.

What carries the argument

The central mechanism is the location-regularised positive pair: instead of two augmented crops of the same image, training views are two separate images taken within a horizontal distance r_loc (Eq. 1), under the proximity assumption that physically close seafloor is more likely to share substrate and habitat class. This substitution is applied uniformly to five CNN-based SSL losses (SimCLR, SimSiam, MoCo-v2, SwAV, DeepCluster-v2) and one ViT self-distillation loss (DINO), yielding the Geo-prefixed variants. The object carrying the argument is r_loc, the radius that controls which images count as positive; set per dataset to approximately the survey image interval (4.0, 0.5, and 2.0 m), it

What would settle it

Measure, on a densely labelled site, the fraction of image pairs within the chosen r_loc (0.5–4.0 m) that actually fall on different semantic classes. If that fraction is large—say, more than a few percent of pairs—the proximity assumption fails, and location-regularised SSL should no longer beat standard SSL; repeating the paper's GeoCLR comparison on such a site while tracking this fraction would settle the claim.

Watch

Extended reading notes

Core claim

The paper's central claim is that location-regularised self-supervised learning—where positive training pairs are two distinct images whose geographic coordinates differ by less than a radius r_loc rather than two augmented views of the same image—consistently improves downstream macro-F1 classification across six SSL strategies and three diverse seafloor datasets. Concretely, average F1 gains over the corresponding standard SSL are 4.9±4.0% for CNN backbones (ResNet18) and 6.3±8.9% for ViT backbones (DINO, fine-tuned to different depths). The best CNN is a low-dimensional location-regularised model (F1=0.778±0.122), exceeding the best pretrained CNN; for ViTs, however, the pretrained high-d

Load-bearing premise

The load-bearing premise is the proximity assumption: seafloor images taken within a few metres of each other almost always show the same substrate or habitat class, so using them as positive training pairs teaches useful features.

Editorial extensions

If this is right

  • Location-regularised SSL is consistently better than its unregularised SSL counterpart across all six methods and three datasets, so the benefit is not specific to the original SimCLR-based GeoCLR.
  • When latent representations are reduced to 128 dimensions, location regularisation matters most for CNNs: it closes much of the gap to pretrained models, making dataset-specific SSL practical on compact features.
  • For ViTs, a pretrained ImageNet-DINO model transfers to seafloor imagery without dataset-specific SSL training, matching the best fine-tuned location-regularised model; compute-limited deployments can skip ViT fine-tuning entirely.
  • Pair-based positive views (SimCLR, SimSiam, MoCo-v2) outperform multi-crop cluster-level methods when location-regularised, because local crops can carry semantically ambiguous content across image boundaries.
  • High-dimensional latent spaces remain beneficial for ViTs, while CNNs are roughly indifferent to dimensionality after location regularisation, indicating the regulariser is doing the work that extra dimensions would otherwise do.

Reading between the lines

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

  • If the proximity assumption holds as well at habitat boundaries as in habitat interiors, an adaptive r_loc—set by local habitat patch size or navigation uncertainty rather than the survey image interval—should improve on the fixed radii used here; this is not tested in the paper.
  • The same location-regularised features should transfer to other self-supervised uses the paper does not evaluate, such as unsupervised clustering for habitat mapping or content-based image retrieval, because the mechanism enforces patch-scale invariance without labels.
  • A deployment-minded extension the authors leave implicit: for memory-constrained AUVs, a low-dimensional location-regularised CNN can substitute for a much larger pretrained ViT with only a small F1 penalty, but the paper does not quantify that computational trade-off.
  • The reported ± values are cross-dataset scatter from single training runs, not run-to-run uncertainty; repeated-seed experiments on one dataset would be a natural follow-up to tighten the ViT gain estimate.
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. This paper studies whether location regularisation—replacing same-image augmentation pairs with physically nearby images as positive pairs—improves self-supervised feature learning for seafloor imagery. It evaluates six SSL methods (SimCLR, SimSiam, MoCo-v2, SwAV, DeepCluster-v2, DINO) and their location-regularised variants across three AUV datasets with CNN and ViT backbones, at high and low latent dimensions. Downstream macro-F1 of SVM classifiers trained on frozen features is reported. The paper claims that location regularisation consistently improves SSL (average F1 gains of 4.9±4.0% for CNNs and 6.3±8.9% for ViTs), that low-dimensional GeoCLR is the best CNN (0.778), and that pretrained DINO ViTs match the best location-regularised fine-tuned ViT (0.795).

Significance. If the results hold, location metadata is a cheap and reliable regulariser for seafloor SSL, and dataset-specific SSL training for ViT-based pipelines may be unnecessary when a pretrained DINO model is available. The study's empirical breadth is a genuine strength: six modern SSL strategies, three independently collected seafloor datasets, controlled backbone and evaluation protocols, and a clearly defined downstream task. The paper also makes a falsifiable claim about pretrained ViT generalisation that is interesting regardless of the SSL comparison. The main weakness is statistical: every reported F1 is a single training-run result, and the train/test split for the downstream SVM is random rather than spatially separated, so the headline claim of 'consistent improvement' is currently under-supported.

major comments (3)
  1. [§IV-B and Tables III–IV] The headline claim that location-regularisation 'consistently improves' SSL rests on per-cell F1 differences that are each measured from a single training run. The paper does not state the number of seeds, and no repeated SSL or downstream-SVM runs are reported; the ± values in the average rows are standard deviations across the three datasets, not over runs. Several gains are small enough to be within stochastic variation: e.g., Table III low-dim SH GeoSwAV is 0.534 vs 0.531 for SwAV, and low-dim CB GeoDeepCluster-v2 is 0.836 vs 0.833; Table IV low-dim SH GeoFT12B is 0.578 vs 0.603 for FT12B, i.e., negative. Without repeated runs or a paired statistical test, the central claim is not statistically supported. Please provide multiple seeds per configuration and report per-cell variance or a significance test.
  2. [§IV-A/B, train/test split] The downstream SVM is trained on a random subset of the labelled images and evaluated on the remaining images, with no spatial separation. Because the survey design produces strong spatial autocorrelation—indeed, the proximity assumption states that nearby images are more similar—random splits can leak location information between training and test examples, inflating F1 and differentially benefiting methods that encode locality. For a fair evaluation of feature generalisation, validation should be spatially disjoint (e.g., held-out survey lines or block CV), or the paper should demonstrate that the main comparisons survive after removing test images within r_loc of the SVM training set.
  3. [§III, Eq. (1)] All location variants depend on the single radius r_loc chosen per dataset (4.0, 0.5, 2.0 m), but no sensitivity analysis is performed here; the paper relies on a prior robustness study. Given that the measured gains are often a few percentage points, the choice of r_loc is load-bearing for the generalisation claim. At a minimum, report how F1 varies over a range of r_loc for one dataset/method, or clearly delimit the claim to the chosen radii. Also clarify whether local views in cluster-level methods are always sampled from one of the two global images; if so, the location regularisation is applied only to the global view pair, which should be stated and justified.
minor comments (4)
  1. [Abstract and Conclusion] '4.9±4.0%' and '6.3±8.9%' should state whether these are relative improvements or absolute percentage-point changes in F1; the current phrasing is ambiguous.
  2. [§IV-B and Table III] Notation is inconsistent: GeoDeepCluster_v2 in §IV-B vs GeoDeepCluster-v2 in Table III, and 'SWaV' appears in several forms. Use one consistent naming throughout.
  3. [§IV-B, ViT fine-tuning] FT2B/FT4B/FT8B/FT12B nomenclature is ambiguous: 'freezing different numbers of transformer blocks (2, 4, 8, and 12)' and the later phrase 'shallow fine-tuned model (GeoFT2B)' are inconsistent unless clarified. Specify whether the number denotes frozen blocks or trainable blocks.
  4. [Reproducibility] No code or data availability statement is given for the implementation. Please add one, including the training/validation split and any random seeds used, to support reproducibility of the comparisons.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's central claims are empirical F1-score comparisons, not derivations from inputs.

full rationale

The paper's central claims — that location-regularisation improves SSL F1-scores and that pre-trained ViTs match location-regularised ViTs — are supported by held-out macro-F1 measurements across three datasets. Location-regularised variants differ from their base SSL methods only in how positive views are sampled (Eq. 1), and the downstream classifiers are trained on human-validated labels. Nothing in the SSL loss functions (Eqs. 3–10) or the evaluation protocol forces the reported improvements; the F1 scores are measured, not fitted. The r_loc values are chosen to match survey image intervals (4.0, 0.5, 2.0 m), not fitted to maximise F1, and the robustness claim about r_loc is a secondary, explicitly cited prior result [26], not the derivation of the main result. The proximity assumption is stated as a premise, and the paper tests its consequences empirically; that is not circular. References to the authors' earlier GeoCLR work [5] are motivational and contextual, while the current study's experiments stand independently. Other self-citations ([9], [26], [27]) support auxiliary claims about label convergence and r_loc robustness, but none is load-bearing for the abstract's quantitative conclusions. The single-run-per-cell protocol is a legitimate statistical robustness concern, but it is not a form of circularity under the enumerated patterns.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The central claim depends on one per-dataset hand-set radius (r_loc), a dimensionality design point, a frozen-block sweep for ViTs, and six domain/background assumptions, chiefly the proximity assumption. Training hyperparameters (epochs, batch size, learning rate, cluster count) are not stated and are silently inherited from the cited SSL implementations. No invented entities: the proximity assumption is a modeling premise, not a postulated object, and the paper introduces no new force, particle, dimension, or conserved quantity.

free parameters (5)
  • r_loc positive-pair radius (DM) = 4.0 m
    Hand-set in Section IV.B by rounding the 3.6 m image patch interval to the nearest 0.5 m; controls which nearby images form positive pairs in Eq. 1.
  • r_loc positive-pair radius (CB) = 0.5 m
    Hand-set in Section IV.B by rounding the 0.5 m image interval; per-dataset choice.
  • r_loc positive-pair radius (SH) = 2.0 m
    Hand-set in Section IV.B by rounding the 1.8 m image interval; per-dataset choice.
  • PCA reduction to 128 latent dimensions = 128
    Design point selected in Section IV.B to test low-dimensional efficiency against default 512 (CNN) and 768 (ViT) features.
  • ViT frozen-block fine-tuning depth = 2, 4, 8, or 12 blocks
    Hyperparameter sweep in Section IV.B; 'best-performing' claims select from this sweep and the best depth differs by dimensionality.
assumptions (6)
  • domain assumption Proximity assumption: physically close locations have similar seafloor characteristics
    Stated in Section III and used by every location-regularised method via Eq. 1; the whole method tests this premise.
  • domain assumption The vast majority of nearby images represent the same semantic group
    Operational version of the proximity assumption in Section III; breaks down in transition regions between habitat types.
  • domain assumption Above 200 images per class is sufficient for downstream SVM training
    Invoked in Section IV.A to justify differing label-set sizes; supported only by the authors' own refs [9], [27].
  • domain assumption PCA to 128 dimensions preserves task-relevant structure
    Used for all low-dimensional comparisons in Section IV.B, citing refs [28], [29] without site-specific validation.
  • standard math Standard SSL losses and collapse-prevention mechanisms behave as in the original papers
    All six baselines are taken as-is from refs [3], [6], [8], [11]-[13]; only positive-pair sampling is modified.
  • domain assumption ImageNet-pretrained ResNet18 and DINO features transfer to seafloor imagery
    Underlies the pre-trained baselines in Tables III and IV and the ViT initialization; transfer is asserted from refs [8], [23].

how reviews work

0 comments
Cite this review

Pith. "Pith review of Investigating Location-Regularised Self-Supervised Feature Learning for Seafloor Visual Imagery." pith.science (2026). https://pith.science/paper/DSD3BKHT

@misc{pith2026250906660,
  author       = {Pith},
  title        = {Pith review of: Investigating Location-Regularised Self-Supervised Feature Learning for Seafloor Visual Imagery},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DSD3BKHT}},
  note         = {Machine review of arXiv:2509.06660}
}
abstract

High-throughput interpretation of robotically gathered seafloor visual imagery can increase the efficiency of marine monitoring and exploration. Although recent research has suggested that location metadata can enhance self-supervised feature learning (SSL), its benefits across different SSL strategies, models and seafloor image datasets are underexplored. This study evaluates the impact of location-based regularisation on six state-of-the-art SSL frameworks, which include Convolutional Neural Network (CNN) and Vision Transformer (ViT) models with varying latent-space dimensionality. Evaluation across three diverse seafloor image datasets finds that location-regularisation consistently improves downstream classification performance over standard SSL, with average F1-score gains of $4.9 \pm 4.0%$ for CNNs and $6.3 \pm 8.9%$ for ViTs, respectively. While CNNs pretrained on generic datasets benefit from high-dimensional latent representations, dataset-optimised SSL achieves similar performance across the high (512) and low (128) dimensional latent representations. Location-regularised SSL improves CNN performance over pre-trained models by $2.7 \pm 2.7%$ and $10.1 \pm 9.4%$ for high and low-dimensional latent representations, respectively. For ViTs, high-dimensionality benefits both pre-trained and dataset-optimised SSL. Although location-regularisation improves SSL performance compared to standard SSL methods, pre-trained ViTs show strong generalisation, matching the best-performing location-regularised SSL with F1-scores of $0.795 \pm 0.075$ and $0.795 \pm 0.077$, respectively. The findings highlight the value of location metadata for SSL regularisation, particularly when using low-dimensional latent representations, and demonstrate strong generalisation of high-dimensional ViTs for seafloor image analysis.

Figures

Figures reproduced from arXiv: 2509.06660 by the authors.

Figure 1
Figure 1. Location regularisation in contrastive SSL. Instead of [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 3
Figure 3. Constructing the SimCLR similarity matrix for a batch [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. The datasets used in this study consist of more than 90k labelled images. These were obtained at various sites using [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figures from the paper (1 more)
Figure 5
Figure 5. Figure 5: Confusion matrices of the top-performing CNN (a) and ViT (b) models, where the left and right of each subfigure show [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

30 extracted references · 23 canonical work pages

  1. [1]

    Monitoring of benthic reference sites: Using an autonomous underwater vehicle,

    S. B. Williams, O. R. Pizarro, M. V . Jakuba, C. R. Johnson, N. S. Barrett, R. C. Babcock, G. A. Kendrick, P. D. Steinberg, A. J. Heyward, P. J. Do- herty, I. Mahon, M. Johnson-Roberson, D. Steinberg, and A. Friedman, “Monitoring of benthic reference sites: Using an autonomous underwater vehicle,”IEEE Robotics & Automation Magazine, vol. 19, no. 1, pp. 73...

    2012

  2. [2]

    A survey on contrastive self-supervised learning,

    A. Jaiswal, A. R. Babu, M. Z. Zadeh, D. Banerjee, and F. Makedon, “A survey on contrastive self-supervised learning,”Technologies, vol. 9, no. 1, p. 2, 2021

    work page 2021

  3. [3]

    A simple framework for contrastive learning of visual representations,

    T. Chen, S. Kornblith, M. Norouzi, and G. Hinton, “A simple framework for contrastive learning of visual representations,” inProceedings of the 37th International Conference on Machine Learning, 2020, pp. 1597– 1607

    work page 2020

  4. [4]

    ugel-Bennett, and B. Thornton, “Learning features from georeferenced seafloor imagery with location guided autoen- coders,

    T. Yamada, A. Pr"ugel-Bennett, and B. Thornton, “Learning features from georeferenced seafloor imagery with location guided autoen- coders,”Journal of Field Robotics, vol. 38, no. 1, pp. 52–67, 2021

    work page 2021

  5. [5]

    ugel-Bennett, S. B. Williams, O. Pizarro, and B. Thornton, “Geoclr: Georeference contrastive learning for efficient seafloor image interpretation,

    T. Yamada, A. Pr"ugel-Bennett, S. B. Williams, O. Pizarro, and B. Thornton, “Geoclr: Georeference contrastive learning for efficient seafloor image interpretation,”Field Robotics, vol. 2, pp. 1134–1155, 2022

    work page 2022

  6. [6]

    Momentum contrast for unsupervised visual representation learning,

    K. He, H. Fan, Y . Wu, S. Xie, and R. Girshick, “Momentum contrast for unsupervised visual representation learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 9729–9738

    work page 2020

  7. [7]

    Self-supervised learning with data augmentations provably isolates content from style,

    J. von Kügelgen, Y . Sharma, L. Gresele, W. Brendel, B. Schölkopf, M. Besserve, and F. Locatello, “Self-supervised learning with data augmentations provably isolates content from style,” inAdvances in Neural Information Processing Systems, vol. 34, 2021, pp. 16 451– 16 467

    work page 2021

  8. [8]

    Emerging properties in self-supervised vision transformers,

    M. Caron, H. Touvron, I. Misra, H. Jégou, J. Mairal, P. Bojanowski, and A. Joulin, “Emerging properties in self-supervised vision transformers,” Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pp. 9650–9660, 2021

    work page 2021

Show all 30 references
  1. [9]

    Guiding labelling effort for efficient learn- ing with georeferenced images,

    T. Yamada, M. Massot-Campos, A. Prügel-Bennett, O. Pizarro, S. B. Williams, and B. Thornton, “Guiding labelling effort for efficient learn- ing with georeferenced images,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, no. 1, pp. 593–607, 2023

    2023

  2. [10]

    Assessing the repeatability of automated seafloor classification algorithms, with application in marine protected area monitoring,

    A. Zelada Leon, V . A. Huvenne, N. M. Benoist, M. Ferguson, B. J. Bett, and R. B. Wynn, “Assessing the repeatability of automated seafloor classification algorithms, with application in marine protected area monitoring,”Remote Sensing, vol. 12, no. 10, 2020. [Online]. Availabl...

    2020

  3. [11]

    Exploring simple siamese representation learning,

    X. Chen and K. He, “Exploring simple siamese representation learning,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2021, pp. 15 750–15 758

    2021

  4. [12]

    Unsupervised learning of visual features by contrasting cluster assign- ments,

    M. Caron, I. Misra, J. Mairal, P. Goyal, P. Bojanowski, and A. Joulin, “Unsupervised learning of visual features by contrasting cluster assign- ments,” 2021

    2021

  5. [13]

    Deep clustering for unsupervised learning of visual features,

    M. Caron, P. Bojanowski, A. Joulin, and M. Douze, “Deep clustering for unsupervised learning of visual features,” inProceedings of the European Conference on Computer Vision, 2018, pp. 132–149

    2018

  6. [14]

    A survey on self-supervised learning: Algorithms, applications, and future trends,

    J. Gui, T. Chen, J. Zhang, Q. Cao, Z. Sun, H. Luo, and D. Tao, “A survey on self-supervised learning: Algorithms, applications, and future trends,” IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024, early Access

    2024

  7. [15]

    BERT: Pre- training of deep bidirectional transformers for language understanding,

    J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “BERT: Pre- training of deep bidirectional transformers for language understanding,” inProceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologi...

    2019

  8. [16]

    Gpt-4 technical report,

    OpenAI, “Gpt-4 technical report,” 2024. [Online]. Available: https: //arxiv.org/abs/2303.08774

    2024 arXiv

  9. [17]

    Self-supervised learning: Generative or contrastive,

    X. Liu, F. Zhang, Z. Hou, L. Mian, Z. Wang, J. Zhang, and J. Tang, “Self-supervised learning: Generative or contrastive,”IEEE Transactions on Knowledge and Data Engineering, vol. 35, no. 1, pp. 857–876, 2023

    2023

  10. [18]

    Advancing surface defect detection: A review of self- supervised learning approaches,

    P. A. Bhaskarrao, “Advancing surface defect detection: A review of self- supervised learning approaches,”International Journal of Engineering Development and Research, vol. 13, no. 2, pp. 134–140, 2025

    2025

  11. [19]

    Concerning nonnegative matrices and doubly stochastic matrices,

    R. Sinkhorn and P. Knopp, “Concerning nonnegative matrices and doubly stochastic matrices,”Pacific Journal of Mathematics, vol. 21, pp. 343–348, 1967. [Online]. Available: https://api.semanticscholar.org/ CorpusID:50329347

    1967

  12. [20]

    Densedino: Boosting dense self- supervised learning with token-based point-level consistency,

    Y . Yuan, X. Fu, Y . Yu, and X. Li, “Densedino: Boosting dense self- supervised learning with token-based point-level consistency,”arXiv preprint arXiv:2306.04654, 2023

    2023 arXiv

  13. [21]

    Bootstrap your own latent: A new approach to self-supervised learning,

    J.-B. Grill, F. Strub, F. Altché, C. Tallec, P. H. Richemond, E. Buchatskaya, C. Doersch, B. Avila Pires, Z. D. Guo, M. G. Azar, B. Piot, K. Kavukcuoglu, R. Munos, and M. Valko, “Bootstrap your own latent: A new approach to self-supervised learning,” inAdvances in Neural Infor...

    2020

  14. [22]

    Transformers in vision: A survey,

    S. Khan, M. Naseer, M. Hayat, S. W. Zamir, F. S. Khan, and M. Shah, “Transformers in vision: A survey,”ACM Computing Surveys, vol. 54, no. 10, pp. 1–41, 2022

    2022

  15. [23]

    Dinov2: Learning robust visual features without supervision,

    M. Oquab, T. Darcet, T. Moutakanni, H. V . V o, M. Szafraniec, V . Khali- dov, P. Fernandez, D. Haziza, F. Massa, A. El-Nouby, M. Assran, N. Ballas, W. Galuba, R. Howes, P.-Y . Huang, S.-W. Li, I. Misra, M. Rabbat, V . Sharma, G. Synnaeve, H. Xu, H. Jégou, J. Mairal, P. Labatu...

    2023 arXiv

  16. [24]

    Unsupervised learning of dense visual representations,

    P. O. Pinheiro, A. Almahairi, R. Y . Benmalek, F. Golemo, and A. C. Courville, “Unsupervised learning of dense visual representations,” in Advances in Neural Information Processing Systems, vol. 33, 2020, pp. 4489–4500

    2020

  17. [25]

    Generation and visualization of large-scale three-dimensional reconstructions from underwater robotic surveys,

    M. Johnson-Roberson, O. Pizarro, S. B. Williams, and I. Mahon, “Generation and visualization of large-scale three-dimensional reconstructions from underwater robotic surveys,”Journal of Field Robotics, vol. 27, no. 1, pp. 21–51, 2010. [Online]. Available: https://onlinelibrary...

    2010 doi

  18. [26]

    Self-supervised learning with multimodal remote sensed maps for seafloor visual class inference,

    C. Liang, J. Cappelletto, M. Massot-Campos, A. Bodenmann, V . A. Huvenne, C. Wardell, B. J. Bett, D. Newborough, and B. Thornton, “Self-supervised learning with multimodal remote sensed maps for seafloor visual class inference,”The International Journal of Robotics Research, p...

    2025

  19. [27]

    Leveraging spatial metadata in machine learning for improved objective quantification of geological drill core,

    L. J. C. Grant, M. Massot-Campos, R. M. Coggon, B. Thornton, F. C. Rotondo, M. Harris, A. D. Evans, and D. A. H. Teagle, “Leveraging spatial metadata in machine learning for improved objective quantification of geological drill core,”Earth and Space Science, vol. 11, no. 3, p....

    2024 doi

  20. [28]

    3D CNN-PCA: A deep-learning-based parameterization for complex geomodels,

    Y . Liu and L. J. Durlofsky, “3D CNN-PCA: A deep-learning-based parameterization for complex geomodels,”Computers & Geosciences, vol. 148, p. 104676, 2021

    2021

  21. [29]

    Compact DINO- ViT: Feature reduction for visual transformer,

    D. R. Chandranegara, P. Niedziela, and B. Cyganek, “Compact DINO- ViT: Feature reduction for visual transformer,”Electronics, vol. 13, no. 23, 2024

    2024

  22. [30]

    Pytorch: An imperative style, high-performance deep learning library,

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antigaet al., “Pytorch: An imperative style, high-performance deep learning library,” inAdvances in Neural Information Processing Systems, vol. 32, 2019

    2019

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.