Pith. sign in

REVIEW 5 major objections 5 minor 40 references

Self-Organizing Visual Prototypes for Non-Parametric Representation Learning

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Pith's one-line read Replacing learned prototype vectors with memory-based nearest-neighbour regions yields state-of-the-art self-supervised visual features.

desk verdict A solid, clearly written non-parametric SSL recipe whose main innovation—k-NN support embeddings—never gets the control ablation that would test whether it is actually doing the work. read the letter →

arxiv 2505.21533 v1 pith:42MOIAWG submitted 2025-05-23 cs.CV cs.LG

classification cs.CVcs.LG
keywords self-supervisedlearningnon-parametricrepresentationsupportembeddingsmaskedimagemodelingretrievalvisiontransformersprototypeFIFOmemory
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that the standard prototypical recipe for self-supervised visual learning—many learned prototype vectors, regularized with centering or Sinkhorn-Knopp—is suboptimal, and replaces it with a non-parametric alternative. A prototype becomes a local region in the encoder's latent space, represented by an anchor embedding drawn from a FIFO memory together with its nearest neighbours, called support embeddings. From this, the paper derives two losses: Eq. (2) for class-level [CLS] tokens and Eq. (4) for masked patch reconstruction (SOP-MIM). On ImageNet and transfer benchmarks, SOP matches or exceeds strong parametric baselines such as iBOT and DINO while using far fewer prototypes and no collapse-prevention regularizers, with the largest gains on retrieval benchmarks. The paper's conclusion is that non-parametric SSL, with anchors resampled randomly every iteration, is stable and scales better with encoder size than the parametric approach.

What carries the argument

The central object is the Self-Organizing Prototype (SOP): a set formed by uniformly sampling an anchor representation from a FIFO memory of recent embeddings and taking its k nearest neighbours, called support embeddings (SEs), via spherical k-NN. The pseudo-dataset $D$ stacks anchors and supports, and $Y$ stores soft contributions equal to each support's cosine similarity to its anchor, so that $P(u) = \sigma(\langle u, D^T \rangle) Y$ is a weighted vote of the region. Equation (2) applies this to [CLS] tokens across views; Eq. (4) applies it to patch tokens for the SOP-MIM task, with patch-level anchors sampled from a separate memory. Random anchor resampling each iteration is what prevents collapse: keeping anchors fixed leads to training collapse, per the paper's own ablation.

What would settle it

Measure the semantic consistency of SOP supports during pre-training: at epochs 1, 5, and 20, compute the average ImageNet class agreement between each anchor and its k support embeddings in the memory. If the neighbourhood-semantics premise is load-bearing, early agreement should be near chance and should rise as k-NN accuracy rises; if strong final accuracy can coexist with chance-level support agreement at the time the losses are computed, the mechanism is not what the paper claims. A complementary test: replace the k nearest neighbours with k randomly drawn memory embeddings of similar average distance; if performance barely changes, locality is not carrying the result.

Watch

Extended reading notes

Core claim

The central claim is that a learnable prototype is a brittle representation for self-supervised clustering: one vector cannot encode all complementary features of a latent region, and over-clustering with large K biases training toward simpler features. SOP replaces each prototype with a Self-Organizing Prototype—an anchor embedding plus its k spherical nearest neighbours in a memory of recent embeddings—so the similarity between a view and a region is a soft weighted vote over several support embeddings, with weights given by cosine similarity to the anchor. The paper introduces non-parametric versions of the two standard SSL losses: a class-level cross-entropy over support-embedding similarities (Eq. 2) and a masked-image-modeling task in which corrupted patch embeddings are reconstructed against patch-level support embeddings (Eq. 4). Empirically, these losses train stable encoders without centering or Sinkhorn-Knopp regularizers, improve over iBOT on k-NN (79.2 vs 78.0 top-1 on ViT-L) and linear probing, and reach state-of-the-art retrieval mAP on ROxford and RParis; gains widen as the backbone grows.

Load-bearing premise

The load-bearing premise is that the k nearest neighbours of an anchor in the memory of recent embeddings are semantically similar enough to describe their region of feature space; early in training, embeddings are nearly random, so those neighbourhoods may be arbitrary.

Editorial extensions

If this is right

  • State-of-the-art k-NN retrieval on ROxford and RParis can be obtained from frozen off-the-shelf features, with mAP gains up to +3.2 over prior SSL methods on hard splits.
  • SOP's performance gap over parametric baselines grows with backbone size, so non-parametric prototypes may be a better scaling recipe for self-supervised ViTs.
  • The SOP-MIM pretext task alone improves k-NN accuracy by 7.3 points over iBOT's MIM, indicating that non-parametric patch tokenization is a strong reconstruction target.
  • The method is robust to the number of SOPs: 1024 SOPs already beat iBOT, and 4096 is near-optimal, reducing the need for over-clustering.
  • No centering or Sinkhorn regularizers are needed; random anchor resampling alone prevents collapse.

Reading between the lines

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

  • Inference: since SOPs are built from a memory of recent embeddings, the method couples optimization to the encoder's own trajectory, which may create a curriculum where early random neighbourhoods become semantic as the encoder sharpens; this could be tested by measuring the class purity of supports over training.
  • Inference: the same non-parametric region-vote mechanism could transfer to other modalities or to semi-supervised settings where a few labels reweight $Y$, though the paper does not test these uses.
  • Inference: if support-embedding locality is the real driver, performance should degrade as memory size shrinks; the reported inverse-U memory curves are consistent with that and give a direct handle on when the mechanism fails.
  • Inference: because each embedding can belong to several SOPs at once, the loss implicitly smooths the feature space, which may explain the robustness gains to background changes; this is an unstated property we draw from the method.
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 / 5 minor

Summary. The paper proposes Self-Organizing Visual Prototypes (SOP), a non-parametric self-supervised learning method for vision transformers. Instead of learning trainable prototypes, SOP samples random anchor embeddings from FIFO memories, uses spherical k-nearest neighbors to collect support embeddings, and optimizes two losses: a global [CLS]-token loss (Eq. 2) and a masked image modeling loss (Eq. 4) based on patch-level support embeddings. The authors claim that SOP achieves state-of-the-art retrieval performance, scales with encoder size, and avoids collapse without centering or Sinkhorn regularizers. The paper provides extensive experiments on ImageNet linear/k-NN probing, semi-supervised fine-tuning, COCO detection/segmentation, ADE20k segmentation, transfer learning, retrieval, and robustness, along with ablations on memory size, number of anchors, number of support embeddings, and anchor selection strategy.

Significance. If the claims are supported, SOP is a practically valuable and conceptually interesting alternative to parametric prototypical SSL: it is clearly specified, avoids learned prototypes, and is tested across a broad benchmark suite. The manuscript is unusually transparent in shipping PyTorch-style pseudo-code, and the ablations cover many design choices (memory size, anchor count, SE count, momentum, masking strategy, anchor resampling). The strongest evidence is the consistent improvement in ImageNet k-NN and retrieval benchmarks over iBOT and DINO, and the scaling behavior with larger encoders. However, the central mechanistic claim — that semantically coherent k-NN support embeddings are what drive the gains — is not directly tested, and several reported improvements are small and lack repeated-seed uncertainty. The paper is therefore a solid empirical contribution that requires additional controls before its conceptual claims can be accepted.

major comments (5)
  1. [Sec. 2.2.1 and Tables C.7-C.9] The causal role of k-NN-selected support embeddings is not tested. The paper assumes that embeddings in a spherical neighborhood share semantic characteristics, and Table C.7 only contrasts fixed anchors (collapse) with random anchors (no collapse); it does not compare k-NN-selected SEs with, for example, randomly sampled support embeddings of matched size. Without an ablation that replaces k-NN SEs with random SEs, the benchmark gains of Eqs. (2) and (4) cannot be attributed to 'self-organizing semantic prototypes' rather than to anchor-based soft-label matching across views. Please add such an ablation and, if possible, a direct measure of neighborhood semantic purity (e.g., label agreement among SEs on a subset of the memory).
  2. [Sec. 2.2.1 vs Appendix A.1] The definition of the SE contribution matrix Y is inconsistent between the main text and the pseudo-code. Section 2.2.1 defines each entry y as the cosine similarity between an SE and its anchor, but the Memory.forward code constructs exp_labels as a smoothed one-hot distribution over anchor indices and never sets contributions from cosine similarity. Since Eq. (2) and the reported results depend on Y, the manuscript must either correct the formula to match the code or provide the similarity-based variant and report whether the two variants produce different results.
  3. [Tables 1-6 and B.2-B.3] No error bars, confidence intervals, or repeated-seed results are reported, and several headline differences are very small: +0.1 APm in Table 2, +0.1-0.2 in Table 3, +0.1 Fm in Table 4, and +0.1-0.3 in semi-supervised results. Single-run differences of this size are often within training noise for SSL pre-training. For the claims of 'state-of-the-art' and 'increasing performance gains', please report at least three seeds for the main k-NN, retrieval, and semi-supervised experiments, or clearly state which comparisons are stable.
  4. [Table 10 and over-clustering discussion] The claim that SOP does not require many prototypes is not an apples-to-apples comparison. In the default configuration each SOP contains one anchor plus eight SEs, so 1024 SOPs involve 9,216 stored embeddings and 4096 SOPs involve 36,864 embeddings, whereas the iBOT baseline uses a single prototype per cluster. The observed robustness to the number of SOPs may reflect the larger effective number of support points rather than a genuine advantage of the SOP structure. Please include a prototype-based baseline with a comparable number of parameters or support points, and temper the over-clustering interpretation accordingly.
  5. [Sec. 3.5 and Table 5] The abstract's claim of 'state-of-the-art performance on many retrieval benchmarks' is supported only by comparisons to DINO, iBOT, and MaSSL on ROxford/RParis, plus a supervised retrieval baseline. The DAVIS comparison in Table 4 also mixes architectures (SOP ViT-B vs supervised ViT-S). To support the state-of-the-art claim, the retrieval evaluation should include more recent and retrieval-specific self-supervised baselines (e.g., DINOv2) or the claim should be restricted to the compared methods.
minor comments (5)
  1. [Sec. 3.5] The text says 'apply k-NN classification on the frozen features' for retrieval, but ROxford/RParis mAP is a retrieval protocol; please clarify whether the reported numbers use k-NN matching or cosine-based ranking.
  2. [Table 4] The header row contains a duplicated 'IN-1K IN-1K' entry for the supervised baseline; please fix the formatting.
  3. [Eqs. (2) and (4)] The temperature parameters s_temp and t_temp appear in the pseudo-code but are not shown in the loss equations; please state explicitly where the temperature is absorbed into P^{[CLS]} and P^{patch}.
  4. [Table C.7] The fixed-anchor row is a dash with no quantitative result; reporting a collapse metric (e.g., average pairwise cosine similarity or k-NN accuracy of the collapsed model) would make the ablation more informative.
  5. [Appendix C.8] The t-SNE figures C.2 and C.3 are referenced as qualitative evidence but no numerical support is given; please state what specific property of the visualization supports the claim of better feature organization.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: SOP's claims are benchmark-verified; the k-NN locality premise is an explicit assumption, not a self-referential output.

full rationale

The paper does not derive any predicted quantity from its own inputs by construction. Equations (2) and (4) define training objectives that match student-view similarity distributions to teacher-view distributions over non-parametric memory embeddings; the claims being made (k-NN, linear probing, semi-supervised fine-tuning, detection, segmentation, retrieval, robustness) are evaluated on external benchmarks such as ImageNet, COCO, ADE20k, ROxford/RParis, DAVIS, and ImageNet-9. The memory and EMA-teacher bootstrap is self-referential in the standard SSL sense, but the paper does not present it as an external ground truth; it is a training signal. Section 2.2.1 explicitly labels the locality premise as an assumption: "This approach assumes that embeddings in a vicinity contain enough information to describe their region in the feature space," and it notes that non-semantic SEs "may be noisy, potentially harming the learned features," which is an empirical risk rather than a circular justification. The paper discloses its debt to the authors' prior MaSSL framework ("under a strict configuration, it is equivalent to the framework of Silva et al. (2024)"), but this is provenance, not a load-bearing uniqueness theorem or an imported ansatz; no equation reduces to a fitted parameter renamed as a prediction. The skeptic's concern about whether k-NN neighborhoods are semantically coherent is a causal-identification and correctness question, not a circularity: failure of the assumption would weaken the method empirically without making any claim true by definition.

Assumptions & free parameters 9 free parameters · 6 assumptions · 0 invented entities

The method's central evaluation depends on several hand-chosen hyperparameters, including memory sizes, anchor counts, SE counts, temperatures, momentum, loss weights, and masking strategy. These are tuned on the same ImageNet benchmark used for the main results, but no parameter is fitted to a physical target. The main axioms are SSL-style domain assumptions: views of an image should agree, k-NN in latent space is semantically meaningful, and random anchor resampling prevents collapse. No new physical or mathematical entities are introduced; SOPs and support embeddings are algorithmic constructs defined from existing embeddings.

free parameters (9)
  • Memory sizes NC and Np = NC=65536, Np=8192
    Chosen by ablations on ImageNet k-NN accuracy (Tables C.5 and C.6), showing inverse-U curves that peak at these values.
  • Number of global anchors K = 4096
    Ablated in Table 10; 1024 to 16384 all work, with 4096 giving the best k-NN (70.0).
  • Number of support embeddings per SOP k = 8 for [CLS], 1 for MIM
    Table 8 shows the [CLS] loss benefits from up to 8 SEs, while the MIM loss neither improves nor degrades with more SEs.
  • Number of local anchors Kdot = 512
    Ablated in Table C.3; 512 is the best of 256, 512, and 1024.
  • Embedding dimension d = 256
    Set as the feature dimensionality for both memories, with no ablation reported.
  • Temperature parameters s_temp and t_temp = not reported
    Appear in the pseudo-code but their actual values are not given in the paper.
  • Momentum m schedule = cosine schedule, default m 0.994 to 1
    Ablated in Table C.4; the default is chosen by performance.
  • Loss weights lambda1 and lambda2 = 1 and 1
    Set to 1 by default with no ablation shown.
  • Masking strategy and ratio = blockwise 0.3, random 0.7
    Figure C.1 shows blockwise masking is better; ratios are fixed by the authors.
assumptions (6)
  • domain assumption Views of the same image should map to similar distributions over SOPs.
    This is the core SSL premise used in Eq. (2) and Eq. (4).
  • domain assumption Spherical k-NN in the current latent space identifies semantically related embeddings.
    Used in Section 2.2.1 to build SOPs; the paper states this assumption explicitly but provides no proof or analysis of when it holds.
  • domain assumption The feature memory is a representative sample of the training distribution.
    FIFO memories of size 65536 and 8192 are assumed to contain enough recent and diverse embeddings for the anchor and k-NN selection to be meaningful.
  • ad hoc to paper Random anchor resampling prevents collapse.
    Table C.7 shows that fixed anchors lead to collapse, but no theoretical guarantee is offered; this is a design choice specific to SOP.
  • standard math Softmax and cosine similarity are appropriate scoring functions.
    Used in Eqs. (1) through (4) and in the pseudo-code; standard choices in SSL.
  • domain assumption Teacher-student momentum encoder provides stable targets.
    Standard SSL assumption; Table C.4 ablates the momentum schedule.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Self-Organizing Visual Prototypes for Non-Parametric Representation Learning." pith.science (2026). https://pith.science/paper/42MOIAWG

@misc{pith2026250521533,
  author       = {Pith},
  title        = {Pith review of: Self-Organizing Visual Prototypes for Non-Parametric Representation Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/42MOIAWG}},
  note         = {Machine review of arXiv:2505.21533}
}
read the original abstract

We present Self-Organizing Visual Prototypes (SOP), a new training technique for unsupervised visual feature learning. Unlike existing prototypical self-supervised learning (SSL) methods that rely on a single prototype to encode all relevant features of a hidden cluster in the data, we propose the SOP strategy. In this strategy, a prototype is represented by many semantically similar representations, or support embeddings (SEs), each containing a complementary set of features that together better characterize their region in space and maximize training performance. We reaffirm the feasibility of non-parametric SSL by introducing novel non-parametric adaptations of two loss functions that implement the SOP strategy. Notably, we introduce the SOP Masked Image Modeling (SOP-MIM) task, where masked representations are reconstructed from the perspective of multiple non-parametric local SEs. We comprehensively evaluate the representations learned using the SOP strategy on a range of benchmarks, including retrieval, linear evaluation, fine-tuning, and object detection. Our pre-trained encoders achieve state-of-the-art performance on many retrieval benchmarks and demonstrate increasing performance gains with more complex encoders.

Figures

Figures reproduced from arXiv: 2505.21533 by the authors.

Figure 1
Figure 1. k-NN top-1 accuracy on ImageNet. of pre-defined prototypes that are presumed to represent hidden clusters in the data. When presented with multiple views as input, these methods assume that prototypes must produce consistent predictions across views of the same image during pre-training. To avoid ill-posed solutions, methods enforce an equipartition constraint (Asano et al., 2019) that uniformly assigns a batch of i… view at source ↗
Figure 2
Figure 2. First, we select a set of random anchors A = {ai} K i=0 (colored squares with patterns) from a set of representations kept in memory (gray sphere). Second, each anchor selects k support embeddings (SEs) (colored diamonds) as their nearest neighbors (2 in this illustration). Each anchor ai and their SEs form an SOP, representing a hidden structure within the data (shaded colored region). Note that a given embedding m… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 35 canonical work pages

  1. [1]

    M., Rupprecht, C., and Vedaldi, A

    Asano, Y. M., Rupprecht, C., and Vedaldi, A. S elf-labelling via simultaneous clustering and representation learning. In Inter. Conf. Learn. Represent. ( ICLR ) , 2019

  2. [2]

    M., Rupprecht, C., and Vedaldi, A

    Asano, Y. M., Rupprecht, C., and Vedaldi, A. Self-labelling via simultaneous clustering and representation learning. In Inter. Conf. Learn. Represent. ( ICLR ) , 2020

  3. [3]

    Self-supervised learning from images with a joint-embedding predictive architecture

    Assran, M., Duval, Q., Misra, I., Bojanowski, P., Vincent, P., Rabbat, M., LeCun, Y., and Ballas, N. Self-supervised learning from images with a joint-embedding predictive architecture. In IEEE / CVF Inter. Conf. Comput. Vis. Pattern Recog. ( CVPR ) , pp.\ 15619--15629, 2023

  4. [4]

    BEiT: BERT Pre-Training of Image Transformers

    Bao, H., Dong, L., Piao, S., and Wei, F. BEiT: BERT Pre-Training of Image Transformers . In Inter. Conf. Learn. Represent. ( ICLR ) , 2021

  5. [5]

    Deep clustering for unsupervised learning of visual features

    Caron, M., Bojanowski, P., Joulin, A., and Douze, M. Deep clustering for unsupervised learning of visual features . In European Conf. Comput. Vis. ( ECCV ) , pp.\ 132--149, 2018

  6. [6]

    Unsupervised pre-training of image features on non-curated data

    Caron, M., Bojanowski, P., Mairal, J., and Joulin, A. Unsupervised pre-training of image features on non-curated data . In IEEE Inter. Conf. Comput. Vis. ( ICCV ) , pp.\ 2959--2968, 2019

  7. [7]

    Unsupervised Learning of Visual Features by Contrasting Cluster Assignments

    Caron, M., Misra, I., Mairal, J., Goyal, P., Bojanowski, P., and Joulin, A. Unsupervised Learning of Visual Features by Contrasting Cluster Assignments . In Adv. Neural Inf. Process. Sys. ( NeurIPS ) , 2020

  8. [8]

    Emerging Properties in Self-Supervised Vision Transformers

    Caron, M., Touvron, H., Misra, I., J \'e gou, H., Mairal, J., Bojanowski, P., and Joulin, A. Emerging Properties in Self-Supervised Vision Transformers . In IEEE Inter. Conf. Comput. Vis. ( ICCV ) , pp.\ 9650--9660, 2021

Show all 40 references
  1. [9]

    A simple framework for contrastive learning of visual representations

    Chen, T., Kornblith, S., Norouzi, M., and Hinton, G. A simple framework for contrastive learning of visual representations . In Inter. Conf. Mach. Learn. ( ICML ) , 2020 a

  2. [10]

    Chen, T., Kornblith, S., Swersky, K., Norouzi, M., and Hinton, G. E. Big self-supervised models are strong semi-supervised learners . In Adv. Neural Inf. Process. Sys. ( NeurIPS ) , volume 33, pp.\ 22243--22255, 2020 b

  3. [11]

    An Empirical Study of Training Self-Supervised Vision Transformers

    Chen, X., Xie, S., and He, K. An Empirical Study of Training Self-Supervised Vision Transformers . In IEEE Inter. Conf. Comput. Vis. ( ICCV ) , 2021

  4. [12]

    Sinkhorn distances: Lightspeed computation of optimal transport

    Cuturi, M. Sinkhorn distances: Lightspeed computation of optimal transport. In Adv. Neural Inf. Process. Sys. ( NeurIPS ) , volume 26, pp.\ 2292--2300, 2013

  5. [13]

    BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

    Devlin, J., Chang, M.-W., Lee, K., and Toutanova, K. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding . In Burstein, J., Doran, C., and Solorio, T. (eds.), Conference of the North A merican Chapter of the Association for Computational Linguistic...

  6. [14]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

    Dosovitskiy, A., Beyer, L., Kolesnikov, A., Weissenborn, D., Zhai, X., Unterthiner, T., Dehghani, M., Minderer, M., Heigold, G., and Gelly, S. An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale . In Inter. Conf. Learn. Represent. ( ICLR ) , 2020

  7. [15]

    With a little help from my friends: Nearest-neighbor contrastive learning of visual representations

    Dwibedi, D., Aytar, Y., Tompson, J., Sermanet, P., and Zisserman, A. With a little help from my friends: Nearest-neighbor contrastive learning of visual representations. In IEEE Inter. Conf. Comput. Vis. ( ICCV ) , pp.\ 9588--9597, October 2021

  8. [16]

    Mask R-CNN

    He, K., Gkioxari, G., Doll \'a r, P., and Girshick, R. Mask R-CNN . In IEEE Inter. Conf. Comput. Vis. ( ICCV ) , pp.\ 2961--2969, 2017

  9. [17]

    Momentum contrast for unsupervised visual representation learning

    He, K., Fan, H., Wu, Y., Xie, S., and Girshick, R. Momentum contrast for unsupervised visual representation learning . In IEEE / CVF Inter. Conf. Comput. Vis. Pattern Recog. ( CVPR ) , pp.\ 9729--9738, 2020

  10. [18]

    3d object representations for fine-grained categorization

    Krause, J., Stark, M., Deng, J., and Fei-Fei, L. 3d object representations for fine-grained categorization. In IEEE Inter. Conf. Comput. Vis. ( ICCV ) , pp.\ 554--561, 2013

  11. [19]

    and Hinton, G

    Krizhevsky, A. and Hinton, G. Learning multiple layers of features from tiny images. 2009

  12. [20]

    Efficient Self-supervised Vision Transformers for Representation Learning

    Li, C., Yang, J., Zhang, P., Gao, M., Xiao, B., Dai, X., Yuan, L., and Gao, J. Efficient Self-supervised Vision Transformers for Representation Learning . In Inter. Conf. Learn. Represent. ( ICLR ) , 2022

  13. [21]

    Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Doll \'a r, P., and Zitnick, C. L. Microsoft COCO: Common Objects in Context . In European Conf. Comput. Vis. ( ECCV ) , pp.\ 740--755. Springer, 2014

  14. [22]

    Swin Transformer: Hierarchical Vision Transformer using Shifted Windows

    Liu, Z., Lin, Y., Cao, Y., Hu, H., Wei, Y., Zhang, Z., Lin, S., and Guo, B. Swin Transformer: Hierarchical Vision Transformer using Shifted Windows . In IEEE Inter. Conf. Comput. Vis. ( ICCV ) , pp.\ 10012--10022, 2021

  15. [23]

    and Zisserman, A

    Nilsback, M.-E. and Zisserman, A. Automated flower classification over a large number of classes. In 2008 Sixth Indian conference on computer vision, graphics & image processing, pp.\ 722--729. IEEE, 2008

  16. [24]

    Oord, A. v. d., Li, Y., and Vinyals, O. Representation Learning with Contrastive Predictive Coding . arXiv preprint arXiv:1807.03748, 2018

  17. [25]

    Oquab, M., Darcet, T., Moutakanni, T., Vo, H. V., Szafraniec, M., Khalidov, V., Fernandez, P., Haziza, D., Massa, F., El-Nouby, A., Howes, R., Huang, P.-Y., Xu, H., Sharma, V., Li, S.-W., Galuba, W., Rabbat, M., Assran, M., Ballas, N., Synnaeve, G., Misra, I., Jegou, H., Maira...

  18. [26]

    Revisiting Oxford and Paris: Large-scale image retrieval benchmarking

    Radenovi \'c , F., Iscen, A., Tolias, G., Avrithis, Y., and Chum, O. Revisiting Oxford and Paris: Large-scale image retrieval benchmarking . In IEEE / CVF Inter. Conf. Comput. Vis. Pattern Recog. ( CVPR ) , pp.\ 5706--5715, 2018

  19. [27]

    S., and Souza, C

    Revaud, J., Almaz \'a n, J., Rezende, R. S., and Souza, C. R. d. Learning with average precision: Training image retrieval with a listwise loss. In IEEE Inter. Conf. Comput. Vis. ( ICCV ) , pp.\ 5107--5116, 2019

  20. [28]

    J., Shewmake, C., Bertics, A., Pennec, X., and Miolane, N

    Sanborn, S., Mathe, J., Papillon, M., Buracas, D., Lillemark, H. J., Shewmake, C., Bertics, A., Pennec, X., and Miolane, N. Beyond euclid: An illustrated guide to modern machine learning with geometric, topological, and algebraic structures. arXiv preprint arXiv:2407.09468, 2024

  21. [29]

    and Ram\' rez Rivera, A

    Silva, T. and Ram\' rez Rivera, A. Representation learning via consistent assignment of views to clusters. In IEEE Inter. Symp. Applied Comput. Intell. Inf. ( SACI ) , pp.\ 987–994, 2022. ISBN 9781450387132. doi:10.1145/3477314.3507267

  22. [30]

    Silva, T., Pedrini, H., and Rivera, A. R. Learning from memory: Non-parametric memory augmented self-supervised learning of visual features. In Inter. Conf. Mach. Learn. ( ICML ) , pp.\ 1--17, July 2024

  23. [31]

    Training data-efficient image transformers & distillation through attention

    Touvron, H., Cord, M., Douze, M., Massa, F., Sablayrolles, A., and J \'e gou, H. Training data-efficient image transformers & distillation through attention. In Inter. Conf. Mach. Learn. ( ICML ) , pp.\ 10347--10357. PMLR, 2021

  24. [32]

    Scan: Learning to classify images without labels

    Van Gansbeke, W., Vandenhende, S., Georgoulis, S., Proesmans, M., and Van Gool, L. Scan: Learning to classify images without labels . In European Conf. Comput. Vis. ( ECCV ) , pp.\ 268--285. Springer, 2020

  25. [33]

    The inaturalist species classification and detection dataset

    Van Horn, G., Mac Aodha, O., Song, Y., Cui, Y., Sun, C., Shepard, A., Adam, H., Perona, P., and Belongie, S. The inaturalist species classification and detection dataset. In IEEE / CVF Inter. Conf. Comput. Vis. Pattern Recog. ( CVPR ) , pp.\ 8769--8778, 2018

  26. [34]

    and Isola, P

    Wang, T. and Isola, P. Understanding contrastive representation learning through alignment and uniformity on the hypersphere. In Inter. Conf. Learn. Represent. ( ICLR ) , pp.\ 9929--9939. PMLR, 2020

  27. [35]

    X., and Lin, D

    Wu, Z., Xiong, Y., Yu, S. X., and Lin, D. Unsupervised feature learning via non-parametric instance discrimination . In IEEE / CVF Inter. Conf. Comput. Vis. Pattern Recog. ( CVPR ) , pp.\ 3733--3742, 2018

  28. [36]

    Noise or signal: The role of image backgrounds in object recognition

    Xiao, K., Engstrom, L., Ilyas, A., and Madry, A. Noise or signal: The role of image backgrounds in object recognition. In Inter. Conf. Learn. Represent. ( ICLR ) , 2020

  29. [37]

    Unified Perceptual Parsing for Scene Understanding

    Xiao, T., Liu, Y., Zhou, B., Jiang, Y., and Sun, J. Unified Perceptual Parsing for Scene Understanding . In European Conf. Comput. Vis. ( ECCV ) , pp.\ 418--434, 2018

  30. [38]

    Scene Parsing through ADE20k Dataset

    Zhou, B., Zhao, H., Puig, X., Fidler, S., Barriuso, A., and Torralba, A. Scene Parsing through ADE20k Dataset . In IEEE / CVF Inter. Conf. Comput. Vis. Pattern Recog. ( CVPR ) , pp.\ 633--641, 2017

  31. [39]

    iBOT: Image BERT Pre-Training with Online Tokenizer

    Zhou, J., Wei, C., Wang, H., Shen, W., Xie, C., Yuille, A., and Kong, T. iBOT: Image BERT Pre-Training with Online Tokenizer . In Inter. Conf. Learn. Represent. ( ICLR ) , 2022

  32. [40]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.