Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Semantic Correspondence: Unified Benchmarking and a Strong Baseline

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Fine-tuning two DINOv2 layers sets semantic matching state of the art

desk verdict A genuinely useful survey and controlled benchmark whose headline SOTA claim is contradicted by its own Table XI. read the letter →

arxiv 2505.18060 v4 pith:TR7L6E73 submitted 2025-05-23 cs.CV

Kaiyan Zhang , Xinghui Li , Jingyi Lu , Kai Han This is my paper · ORCID
classification cs.CV
keywords semanticcorrespondencematchingDINOv2featurebackbonecostaggregationfine-tuningbenchmarkingsurvey
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 tries to establish that, for semantic correspondence, the choice of feature backbone and whether it is fine-tuned matter far more than elaborate matching modules. Through controlled experiments on SPair-71k, the authors report that DINOv2 is the most powerful backbone, and that fine-tuning only its last two transformer layers raises PCK@0.1 from 40.0% in zero-shot evaluation to 85.1% after supervised fine-tuning. They then show that appending a ResNet bottleneck or a Neighbourhood Consensus cost aggregator yields state-of-the-art results on SPair-71k and PF-PASCAL, and comparable results on AP-10K. A reader should care because the paper provides a unified benchmark and a simple baseline that most complex published pipelines do not beat.

What carries the argument

The load-bearing object is the DINOv2 self-supervised Vision Transformer feature backbone, adapted by fine-tuning its last two layers under the SimSC learning framework: a cross-entropy loss between each query feature and the whole target feature map, with a learned softmax temperature. Around this backbone the paper places a ResNet bottleneck block for feature enhancement and a light Neighbourhood Consensus 4D convolutional aggregator for cost-volume refinement. The mechanism is that DINOv2 already provides strong semantic features, fine-tuning aligns them to keypoint matching, and the extra modules filter local matching patterns.

What would settle it

Run the same fine-tuning recipe (SimSC loss, adaptive temperature, 840x840 resolution) on iBOT, SD2-1, and DINOv2 and compare PCK@0.1 on SPair-71k; if any non-DINOv2 backbone matches or exceeds DINOv2, the claims that DINOv2 is the most powerful backbone and that backbone fine-tuning is the decisive factor would be overturned.

Watch

Extended reading notes

Core claim

The central claim is that a strong pre-trained feature extractor, adapted by light fine-tuning, dominates semantic correspondence performance, so the field's elaborate cost-aggregation and feature-assembly designs add little once the backbone is fine-tuned. Concretely, the authors report that fine-tuning the last two transformer blocks of DINOv2 at 840x840 resolution achieves PCK@0.1 of 85.1% on SPair-71k (second-best overall), and that adding a ResNet bottleneck or Neighbourhood Consensus aggregator reaches 85.2% while improving PF-PASCAL, making the combined system state of the art. They also assert that DINOv2 is the most powerful feature backbone for semantic correspondence, superior to Stable Diffusion features alone or combined with DINOv2, once fine-tuning is applied.

Load-bearing premise

All fine-tuning conclusions are drawn inside the authors' own SimSC training recipe, and if that recipe favors DINOv2 more than it favors other backbones, the ranking 'DINOv2 is most powerful' could be an artifact of the recipe rather than a property of the backbones.

Editorial extensions

If this is right

  • Fine-tuning a strong self-supervised backbone, rather than inventing a new matching module, is the cheapest route to state-of-the-art semantic correspondence.
  • Complex diffusion-based pipelines such as SD+DINOv2 lose their advantage once DINOv2 alone is fine-tuned, so future work can drop the diffusion branch for supervised settings.
  • The unified comparison tables make it possible to read off which configurations, such as resolution, fine-tuning, and multilayer features, drive reported gains across the literature.
  • The strong result on AP-10K suggests the same recipe transfers to new, larger animal-keypoint benchmarks without dataset-specific engineering.

Reading between the lines

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

  • My inference: much of the recent complexity in semantic correspondence, including attention cost aggregators and diffusion-feature assembly, may be compensating for weak or frozen backbones; once the backbone is fine-tuned, simpler components suffice.
  • My inference: the same two-layer fine-tuning recipe could be tested on other dense correspondence tasks such as part segmentation or image editing, where DINOv2 features are already used zero-shot.
  • My inference: the reported resolution sensitivity of DINOv2 suggests that papers comparing methods at different input resolutions may be measuring resolution effects rather than method effects; the unified table's resolution column should be treated as a confounder in future meta-analyses.
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 / 5 minor

Summary. The paper presents a survey of semantic correspondence methods, a taxonomy, a unified benchmark table of existing methods, and a set of controlled experiments that compare feature backbones, feature enhancement modules, and cost aggregators. Based on those experiments, it proposes a strong baseline built on a fine-tuned DINOv2 backbone, optionally augmented with a ResNet bottleneck or a Neighbourhood Consensus cost aggregator, and claims state-of-the-art results on SPair-71k and PF-PASCAL. The manuscript also releases code. The survey and controlled benchmarking are potentially useful contributions, but the headline empirical claims contain internal inconsistencies that must be resolved before the paper can be accepted.

Significance. If the empirical claims are corrected and supported, the paper would provide a valuable reference: it aggregates a large body of work into a structured taxonomy, introduces a unified comparison table with configuration details, and reports controlled experiments that isolate the effect of backbones, enhancement modules, and aggregators. The public code release and the proposed simple baseline are additional strengths, and the finding that fine-tuning a strong backbone is more impactful than adding matching modules is a practically useful message. However, the current text of Section VI-E is contradicted by the paper's own Table XI, and the transformer-versus-CNN aggregator comparison uses a simplified CATs variant, so the central claims need revision or additional experiments before the significance can be fully assessed.

major comments (3)
  1. [Section VI-E and Table XI] The claim that 'when incorporating the ResNet bottleneck and the Neighbourhood Consensus cost aggregator, our approach attains state-of-the-art results on both SPair-71k and PF-PASCAL' is contradicted by Table XI. For PF-PASCAL PCK@0.1, Ours(DINOv2+ResNet) scores 90.8 and Ours(DINOv2+NC) scores 91.5, both below GeoAware-SC's 95.1; only Ours(DINOv2) at 95.8 exceeds that value. Additionally, the same sentence says fine-tuning the last two DINOv2 layers alone gives 'second-best' performance on SPair-71k, but Table XI shows Ours(DINOv2) at 85.1 PCK@0.1, above GeoAware-SC's 82.9, which is first rather than second. The text and the reported numbers cannot both be correct, and the abstract and contributions repeat the state-of-the-art claim, so the manuscript's central result is currently unsupported.
  2. [Section VI-C and Table A2] The takeaway that 'CNN cost aggregators are more effective than Transformer aggregators' is based on a simplified CATs aggregator: the text states that the CATs aggregator is reduced from six layers to a single layer for computational reasons. Table IX and Table A2 then show CATs performing substantially worse than NeighConsensus, but this comparison does not test the original CATs architecture that appears in the literature. The conclusion about transformer-based cost aggregation should either be backed by experiments with the original six-layer CATs at a feasible resolution or be explicitly restricted to the simplified variant.
  3. [Section VI-A2] The conclusion that 'fine-tuning feature backbones plays a crucial role' and the ranking of DINOv2 as the most powerful backbone are based entirely on the SimSC training framework, which is the authors' own design. All fine-tuning experiments 'follow the learning framework in SimSC [58]' with cross-entropy loss and adaptive temperature, but there is no comparison with an alternative fine-tuning recipe or an ablation showing that this framework is equally suitable for SD2-1, DINOv2, and the combined SD2-1+DINOv2 backbone. Without such evidence, the comparative ranking could reflect properties of the training framework rather than the backbones themselves; at minimum, the claims should be qualified and the framework sensitivity should be discussed.
minor comments (5)
  1. [Table V] The row 'SD2-1+DINOv2 (960,840), 60' lists the PCK values as '47.960.467.0', which appears to be missing delimiters; it should read '47.9 60.4 67.0'.
  2. [Section VI-A2 and Table VII] The phrase 'declines when deeper layers are fine-tuned' is potentially confusing because the table's 'FF.' column denotes the stage from which fine-tuning starts; fine-tuning from stage 10 or 9 means more layers are trained, not necessarily 'deeper' layers. The interpretation would be clearer if the text said 'when more stages are fine-tuned'.
  3. [Table XI] The large degradation on PF-WILLOW for Ours(DINOv2+NC) (62.3 vs. 73.7 PCK@0.1 for Ours(DINOv2)) is not discussed in the text; since Table XI is the central comparison table, the authors should comment on this drop and explain why the NC module helps on some benchmarks but hurts on PF-WILLOW.
  4. [Section VI and Appendix A] The controlled experiments report single numbers without error bars or repeated-seed variance; for differences such as the +2.4 versus -0.8 PCK changes in Table VIII, this makes it difficult to assess whether the observed gaps are significant. Reporting mean and standard deviation over multiple training runs would strengthen the conclusions.
  5. [Section III-C2] The text says parameterized transformation represents matches as 'the wrapping between two images', which appears to be a typo for 'warping'; the same typo also appears in the Figure 6 caption.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's claims rest on external benchmark evaluations and controlled experiments, not on definitional equivalences or load-bearing self-citations.

full rationale

The paper's main empirical claims — DINOv2 is the most powerful backbone and the proposed baseline achieves state-of-the-art results — are tested against external benchmarks (SPair-71k, PF-PASCAL, PF-WILLOW, AP-10K) using standard PCK metrics, and the zero-shot versus fine-tuned comparisons are controlled experiments rather than definitional reductions. The training recipe is cited from the authors' prior SimSC and SD4Match work, but the cited papers are used as methodological references, not as the evidence for the conclusions; the reported numbers come from the paper's own runs. Selecting the best-performing backbone, enhancement module, and cost aggregator from Tables VIII and IX and then presenting that configuration as the baseline is in-sample selection, which is a methodological overfitting concern, not circularity in the sense of a fitted parameter renamed as a prediction or an equation reducing to its own input. The paper does contain an internal inconsistency between the Section VI-E statement that the module-augmented variants achieve state-of-the-art results on both SPair-71k and PF-PASCAL and the values in Table XI, where Ours(DINOv2+ResNet) and Ours(DINOv2+NC) score lower than Ours(DINOv2) on PF-PASCAL; this is a correctness or reporting issue, not a circularity issue. No self-citation chain, uniqueness theorem, or ansatz-importing citation is load-bearing for the central derivation, so no significant circularity is present.

Assumptions & free parameters 4 free parameters · 4 assumptions · 1 invented entities

The central empirical claims rest on a handful of hand-chosen configurations (fine-tuned stage, resolution, aggregator simplifications) plus standard domain assumptions about metrics and dataset representativeness. No new entities are introduced.

free parameters (4)
  • Number of fine-tuned DINOv2 stages = stage 11 (last two blocks)
    Selected by scanning stages 8-12 in Table VII; performance peaks at stage 11, so the final baseline uses this hand-chosen stage count. This is model selection on a validation split, not a fitted constant.
  • Input resolution = 840x840
    The paper evaluates 224, 448, 840 and chooses 840 for the final baseline; higher resolution consistently helps, but the specific value is a design choice.
  • Neighbourhood Consensus filter size and channels = 3x3x3x3 filters, half channels
    The paper modifies NC-Net's original 5x5x5x5 and 16-channel config to reduce compute; this modified aggregator is used in the final SOTA baseline.
  • CATs aggregator layers = 1 layer instead of 6
    The paper simplifies CATs to a single layer in the comparison, which affects the conclusion that CNN aggregators beat transformer aggregators.
assumptions (4)
  • domain assumption The PCK metric with bbox-based thresholds is a valid measure of semantic correspondence quality.
    Used without justification as the evaluation standard; the choice of alpha values (0.05, 0.1, 0.15) and per-image vs per-point averaging affects rankings.
  • domain assumption The evaluated backbones and modules are representative of the semantic correspondence literature.
    The controlled experiments cover six backbones, three feature enhancement modules, and three cost aggregators; the paper assumes these span the design space, which limits the generality of the 'backbone dominates' conclusion.
  • ad hoc to paper SimSC's training framework (cross-entropy loss with adaptive temperature) is uniformly suitable for all backbones.
    The fine-tuning conclusion is established using the authors' own SimSC framework; if this framework is better suited to DINOv2 than to other backbones, the comparisons in Tables VI and VII are biased.
  • domain assumption Aggregated literature results are comparable despite differing resolutions, backbones, and training configurations.
    Tables II-IV merge results from papers with different settings; the paper acknowledges this and adds configuration columns, but the unified SOTA claim still rests on comparability.
invented entities (1)
  • None
    purpose: No new entities are introduced.
    The paper uses existing backbones (DINOv2, Stable Diffusion) and modules (Neighbourhood Consensus, ResNet bottleneck); no new particles, forces, or abstract constructs are postulated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Semantic Correspondence: Unified Benchmarking and a Strong Baseline." pith.science (2026). https://pith.science/paper/TR7L6E73

@misc{pith2026250518060,
  author       = {Pith},
  title        = {Pith review of: Semantic Correspondence: Unified Benchmarking and a Strong Baseline},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/TR7L6E73}},
  note         = {Machine review of arXiv:2505.18060}
}
read the original abstract

Establishing semantic correspondence is a challenging task in computer vision, aiming to match keypoints with the same semantic information across different images. Benefiting from the rapid development of deep learning, remarkable progress has been made over the past decade. However, a comprehensive review and analysis of this task remains absent. In this paper, we present the first extensive survey of semantic correspondence methods. We first propose a taxonomy to classify existing methods based on the type of their method designs. These methods are then categorized accordingly, and we provide a detailed analysis of each approach. Furthermore, we aggregate and summarize the results of methods in literature across various benchmarks into a unified comparative table, with detailed configurations to highlight performance variations. Additionally, to provide a detailed understanding on existing methods for semantic matching, we thoroughly conduct controlled experiments to analyse the effectiveness of the components of different methods. Finally, we propose a simple yet effective baseline that achieves state-of-the-art performance on multiple benchmarks, providing a solid foundation for future research in this field. We hope this survey serves as a comprehensive reference and consolidated baseline for future development. Code is publicly available at: https://github.com/Visual-AI/Semantic-Correspondence.

Figures

Figures reproduced from arXiv: 2505.18060 by the authors.

Figure 1
Figure 1. The development timeline of semantic correspondence methods categorized by their levels of supervision. These categories include strongly supervised, weakly supervised, and zero-shot methods. Each method is labeled with its backbone architecture, indicated by distinct symbols: ⃝ for CNN backbones, ✩ for vision transformer backbones, and • for stable diffusion backbones. For methods that utilize both vision transform… view at source ↗
Figure 2
Figure 2. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Pipeline for feature enhancement methods. The feature extractor generates feature maps F s and F t from the source image I s and target image I t , respectively. After channel-wise L2 normalization, their dot product constructs a cosine similarity matrix (2D correlation map) for each query point in F s . The correlation map is transformed into a probability distribution through a localization operation such as soft-… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Pipeline for cost volume-based methods. The feature extractor generates feature maps F s and F t from the source image I s and target image I t , respectively. After normalization, their dot product constructs a cost volume for each query point in F s , storing the cos…
Figure 5
Figure 5. Figure 5: Pipeline for flow field-based methods. Given a pair of images I s and I t , feature extraction is performed to obtain dense feature maps F s and F t respectively. The cost construction step is then applied to derive a cost volume, which is subsequently transformed into…
Figure 6
Figure 6. Figure 6: Pipeline for parameterized transformation-based methods. Given a pair of images I s and I t , feature extraction is performed to obtain dense feature maps F s and F t respectively. The cost construction step is then applied to derive a cost volume, which is subsequentl…
Figure 7
Figure 7. Figure 7: Performance of different feature backbones on zero-shot evaluation [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]
Figure 8
Figure 8. Figure 8: Performance of ResNet bottleneck and neighbourhood consensus [PITH_FULL_IMAGE:figures/full_fig_p016_8.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Emergent Region-Level Facial Correspondence in Frozen Vision Foundation Models

    cs.CV 2026-07 conditional novelty 6.0 of 10

    Frozen DINOv3 features achieve 83% cross-identity facial region matching and 95.5% temporal region tracking on CelebDF-v2 with no face-specific or temporal training.

Reference graph

Works this paper leans on

138 extracted references · 63 canonical work pages · cited by 1 Pith paper

  1. [58]

    Simsc: A simple framework for semantic correspondence with temperature learning,

    X. Li, K. Han, X. Wan, and V . A. Prisacariu, “Simsc: A simple framework for semantic correspondence with temperature learning,” arXiv preprint arXiv:2305.02385, 2023

    arXiv 2023

  2. [1]

    Sfnet: Learning object-aware semantic correspondence,

    J. Lee, D. Kim, J. Ponce, and B. Ham, “Sfnet: Learning object-aware semantic correspondence,” inCVPR, 2019

    2019

  3. [2]

    Learning semantic correspondence exploiting an object-level prior,

    J. Lee, D. Kim, W. Lee, J. Ponce, and B. Ham, “Learning semantic correspondence exploiting an object-level prior,”TPAMI, 2022

    2022

  4. [3]

    Sift flow: Dense correspondence across scenes and its applications,

    C. Liu, J. Yuen, and A. Torralba, “Sift flow: Dense correspondence across scenes and its applications,”TPAMI, 2011

    2011

  5. [4]

    Reference- based sketch image colorization using augmented-self reference and dense semantic correspondence,

    J. Lee, E. Kim, Y . Lee, D. Kim, J. Chang, and J. Choo, “Reference- based sketch image colorization using augmented-self reference and dense semantic correspondence,” inCVPR, 2020

    2020

  6. [5]

    Tokenflow: Con- sistent diffusion features for consistent video editing,

    M. Geyer, O. Bar-Tal, S. Bagon, and T. Dekel, “Tokenflow: Con- sistent diffusion features for consistent video editing,”arXiv preprint arXiv:2307.10373, 2023

    arXiv 2023

  7. [6]

    Regiondrag: Fast region-based image editing with diffusion models,

    J. Lu, X. Li, and K. Han, “Regiondrag: Fast region-based image editing with diffusion models,” inECCV, 2024

    2024

  8. [7]

    Neural congealing: Aligning images to a joint semantic atlas,

    D. Ofri-Amar, M. Geyer, Y . Kasten, and T. Dekel, “Neural congealing: Aligning images to a joint semantic atlas,” inCVPR, 2023

    2023

Show all 138 references
  1. [8]

    DragonDiffusion: Enabling drag-style manipulation on diffusion models,

    C. Mou, X. Wang, J. Song, Y . Shan, and J. Zhang, “DragonDiffusion: Enabling drag-style manipulation on diffusion models,”arXiv preprint arXiv:2307.02421, 2023

    2023 arXiv

  2. [9]

    Unsupervised object discovery and localization in the wild: Part-based matching with bottom-up region proposals,

    M. Cho, S. Kwak, C. Schmid, and J. Ponce, “Unsupervised object discovery and localization in the wild: Part-based matching with bottom-up region proposals,” inCVPR, 2015

    2015

  3. [10]

    Histograms of oriented gradients for human detection,

    N. Dalal and B. Triggs, “Histograms of oriented gradients for human detection,” inCVPR, 2005. 18

    2005

  4. [11]

    Scnet: Learning semantic correspondence,

    K. Han, R. S. Rezende, B. Ham, K.-Y . K. Wong, M. Cho, C. Schmid, and J. Ponce, “Scnet: Learning semantic correspondence,” inICCV, 2017

    2017

  5. [12]

    Fcss: Fully convolutional self-similarity for dense semantic correspondence,

    S. Kim, D. Min, B. Ham, S. Jeon, S. Lin, and K. Sohn, “Fcss: Fully convolutional self-similarity for dense semantic correspondence,” in CVPR, 2017

    2017

  6. [13]

    Hyperpixel flow: Semantic correspondence with multi-layer neural features,

    J. Min, J. Lee, J. Ponce, and M. Cho, “Hyperpixel flow: Semantic correspondence with multi-layer neural features,” inICCV, 2019

    2019

  7. [14]

    Learning to compose hypercolumns for visual correspondence,

    ——, “Learning to compose hypercolumns for visual correspondence,” inECCV, 2020

    2020

  8. [15]

    Dynamic context correspondence network for semantic alignment,

    S. Huang, Q. Wang, S. Zhang, S. Yan, and X. He, “Dynamic context correspondence network for semantic alignment,” inICCV, 2019

    2019

  9. [16]

    Efficient semantic matching with hypercolumn correlation,

    S. Kim, J. Min, and M. Cho, “Efficient semantic matching with hypercolumn correlation,” inWACV, 2024

    2024

  10. [17]

    Neighbourhood consensus networks,

    I. Rocco, M. Cimpoi, R. Arandjelovi ´c, A. Torii, T. Pajdla, and J. Sivic, “Neighbourhood consensus networks,” inNeurIPS, 2018

    2018

  11. [18]

    Correspondence networks with adaptive neighbourhood consensus,

    S. Li, K. Han, T. W. Costain, H. Howard-Jenkins, and V . Prisacariu, “Correspondence networks with adaptive neighbourhood consensus,” inCVPR, 2020

    2020

  12. [19]

    Convolutional hough matching networks,

    J. Min and M. Cho, “Convolutional hough matching networks,”CVPR, 2021

    2021

  13. [20]

    Patchmatch-based neighborhood consensus for semantic correspondence,

    J. Y . Lee, J. DeGol, V . Fragoso, and S. N. Sinha, “Patchmatch-based neighborhood consensus for semantic correspondence,” inCVPR, 2021

    2021

  14. [21]

    Cats: Cost aggregation transformers for visual correspondence,

    S. Cho, S. Hong, S. Jeon, Y . Lee, K. Sohn, and S. Kim, “Cats: Cost aggregation transformers for visual correspondence,” inNeurIPS, 2021

    2021

  15. [22]

    Cats++: Boosting cost aggregation with convolutions and transformers,

    S. Cho, S. Hong, and S. Kim, “Cats++: Boosting cost aggregation with convolutions and transformers,”TPAMI, 2023

    2023

  16. [23]

    Transformatcher: Match-to-match attention for semantic correspondence,

    S. Kim, J. Min, and M. Cho, “Transformatcher: Match-to-match attention for semantic correspondence,” inCVPR, 2022

    2022

  17. [24]

    DINOv2: Learning robust visual features without supervision,

    M. Oquab, T. Darcet, T. Moutakanni, H. V o, M. Szafraniec, V . Khali- dov, P. Fernandez, D. Haziza, F. Massa, A. El-Noubyet al., “DINOv2: Learning robust visual features without supervision,” arXiv:2304.07193, 2023

    2023 arXiv

  18. [25]

    High- resolution image synthesis with latent diffusion models,

    R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High- resolution image synthesis with latent diffusion models,” inCVPR, 2022

    2022

  19. [26]

    Emergent correspondence from image diffusion,

    L. Tang, M. Jia, Q. Wang, C. P. Phoo, and B. Hariharan, “Emergent correspondence from image diffusion,”NeurIPS, 2023

    2023

  20. [27]

    A tale of two features: Stable diffusion complements dino for zero-shot semantic correspondence,

    J. Zhang, C. Herrmann, J. Hur, L. Polania Cabrera, V . Jampani, D. Sun, and M.-H. Yang, “A tale of two features: Stable diffusion complements dino for zero-shot semantic correspondence,”NeurIPS, 2023

    2023

  21. [28]

    Sd4match: Learning to prompt stable diffusion model for semantic matching,

    X. Li, J. Lu, K. Han, and V . Prisacariu, “Sd4match: Learning to prompt stable diffusion model for semantic matching,”CVPR, 2024

    2024

  22. [29]

    Telling left from right: Identifying geometry-aware semantic correspondence,

    J. Zhang, C. Herrmann, J. Hur, E. Chen, V . Jampani, D. Sun, and M.- H. Yang, “Telling left from right: Identifying geometry-aware semantic correspondence,” inCVPR, 2024

    2024

  23. [30]

    Distillation of diffusion features for semantic correspondence,

    F. Fundel, J. Schusterbauer, V . T. Hu, and B. Ommer, “Distillation of diffusion features for semantic correspondence,” inWACV, 2025

    2025

  24. [31]

    Hartley and A

    R. Hartley and A. Zisserman,Multiple view geometry in computer vision. Cambridge university press, 2003

    2003

  25. [32]

    Image matching from handcrafted to deep features: A survey,

    J. Ma, X. Jiang, A. Fan, J. Jiang, and J. Yan, “Image matching from handcrafted to deep features: A survey,”IJCV, 2021

    2021

  26. [33]

    A computational theory of human stereo vi- sion,

    D. Marr and T. Poggio, “A computational theory of human stereo vi- sion,”Proceedings of the Royal Society of London. Series B. Biological Sciences, 1979

    1979

  27. [34]

    Determining optical flow,

    B. K. Horn and B. G. Schunck, “Determining optical flow,”Artificial intelligence, 1981

    1981

  28. [35]

    Distinctive image features from scale-invariant key- points,

    D. G. Lowe, “Distinctive image features from scale-invariant key- points,”IJCV, 2004

    2004

  29. [36]

    A maximum entropy framework for part-based texture and object recognition,

    S. Lazebnik, C. Schmid, and J. Ponce, “A maximum entropy framework for part-based texture and object recognition,” inICCV’05, 2005

    2005

  30. [37]

    Flexible object models for category-level 3d object recognition,

    A. Kushal, C. Schmid, and J. Ponce, “Flexible object models for category-level 3d object recognition,” inCVPR, 2007

    2007

  31. [38]

    Deformable spatial pyramid matching for fast dense correspondences,

    J. Kim, C. Liu, F. Sha, and K. Grauman, “Deformable spatial pyramid matching for fast dense correspondences,” inCVPR, 2013

    2013

  32. [39]

    Daisy filter flow: A generalized discrete approach to dense correspondences,

    H. Yang, W.-Y . Lin, and J. Lu, “Daisy filter flow: A generalized discrete approach to dense correspondences,” inCVPR, 2014

    2014

  33. [40]

    Daisy: An efficient dense descriptor applied to wide-baseline stereo,

    E. Tola, V . Lepetit, and P. Fua, “Daisy: An efficient dense descriptor applied to wide-baseline stereo,”TPAMI, 2010

    2010

  34. [43]

    Dense semantic correspon- dence where every pixel is a classifier,

    H. Bristow, J. Valmadre, and S. Lucey, “Dense semantic correspon- dence where every pixel is a classifier,” inICCV, 2015

    2015

  35. [44]

    Generalized deformable spatial pyramid: Geometry-preserving dense correspondence estimation,

    J. Hur, H. Lim, C. Park, and S. C. Ahn, “Generalized deformable spatial pyramid: Geometry-preserving dense correspondence estimation,” in CVPR, 2015

    2015

  36. [45]

    Proposal flow,

    B. Ham, M. Cho, C. Schmid, and J. Ponce, “Proposal flow,” inCVPR, 2016

    2016

  37. [46]

    Proposal flow: Semantic correspondences from object propos- als,

    ——, “Proposal flow: Semantic correspondences from object propos- als,”TPAMI, 2018

    2018

  38. [48]

    Object-aware dense semantic correspondence,

    F. Yang, X. Li, H. Cheng, J. Li, and L. Chen, “Object-aware dense semantic correspondence,” inCVPR, 2017

    2017

  39. [49]

    A graph-matching kernel for object categorization,

    O. Duchenne, A. Joulin, and J. Ponce, “A graph-matching kernel for object categorization,” inICCV, 2011

    2011

  40. [50]

    Progressive graph matching: Making a move of graphs via probabilistic voting,

    M. Cho and K. M. Lee, “Progressive graph matching: Making a move of graphs via probabilistic voting,” inCVPR, 2012

    2012

  41. [51]

    Universal corre- spondence network,

    C. Choy, J. Gwak, S. Savarese, and M. Chandraker, “Universal corre- spondence network,” inNeurIPS, 2016

    2016

  42. [52]

    Do convnets learn correspon- dence?

    J. L. Long, N. Zhang, and T. Darrell, “Do convnets learn correspon- dence?” inNeurIPS, 2014

    2014

  43. [53]

    Hypercolumns for object segmentation and fine-grained localization,

    B. Hariharan, P. Arbel ´aez, R. Girshick, and J. Malik, “Hypercolumns for object segmentation and fine-grained localization,” inCVPR, 2015

    2015

  44. [54]

    Multi-scale matching networks for semantic correspondence,

    D. Zhao, Z. Song, Z. Ji, G. Zhao, W. Ge, and Y . Yu, “Multi-scale matching networks for semantic correspondence,” inICCV, 2021

    2021

  45. [55]

    Independently keypoint learning for small object semantic correspondence,

    H. Jin and H. Li, “Independently keypoint learning for small object semantic correspondence,”arXiv preprint arXiv:2404.02678, 2024

    2024 arXiv

  46. [56]

    Pixel-level semantic correspondence through layout-aware represen- tation learning and multi-scale matching integration,

    Y . Sun, Z. Yin, H. Wang, Y . Wang, X. Qiu, W. Ge, and W. Zhang, “Pixel-level semantic correspondence through layout-aware represen- tation learning and multi-scale matching integration,” inCVPR, 2024

    2024

  47. [57]

    Dif- fusion hyperfeatures: Searching through time and space for semantic correspondence,

    G. Luo, L. Dunlap, D. H. Park, A. Holynski, and T. Darrell, “Dif- fusion hyperfeatures: Searching through time and space for semantic correspondence,”NeurIPS, 2023

    2023

  48. [59]

    Unsupervised semantic correspondence using stable diffusion,

    E. Hedlin, G. Sharma, S. Mahajan, H. Isack, A. Kar, A. Tagliasacchi, and K. M. Yi, “Unsupervised semantic correspondence using stable diffusion,”NeurIPS, 2024

    2024

  49. [60]

    Lift: A surprisingly simple lightweight feature transform for dense vit descriptors,

    S. Suri, M. Walmer, K. Gupta, and A. Shrivastava, “Lift: A surprisingly simple lightweight feature transform for dense vit descriptors,” in ECCV, 2024

    2024

  50. [61]

    Speech understanding systems: Report of a steering committee,

    M. Medress, F. Cooper, J. Forgie, C. Green, D. Klatt, M. O’Malley, E. Neuburg, A. Newell, D. Reddy, B. Ritea, J. Shoup-Hummel, D. Walker, and W. Woods, “Speech understanding systems: Report of a steering committee,”Artificial Intelligence, 1977

    1977

  51. [62]

    Feature pyramid networks for object detection,

    T.-Y . Lin, P. Doll´ar, R. Girshick, K. He, B. Hariharan, and S. Belongie, “Feature pyramid networks for object detection,” inCVPR, 2017

    2017

  52. [63]

    An image is worth one word: Personal- izing text-to-image generation using textual inversion,

    R. Gal, Y . Alaluf, Y . Atzmon, O. Patashnik, A. H. Bermano, G. Chechik, and D. Cohen-Or, “An image is worth one word: Personal- izing text-to-image generation using textual inversion,”arXiv preprint arXiv:2208.01618, 2022

    2022 arXiv

  53. [64]

    Zero-shot image feature consensus with deep functional maps,

    X. Cheng, C. Deng, A. W. Harley, Y . Zhu, and L. Guibas, “Zero-shot image feature consensus with deep functional maps,” inECCV, 2024

    2024

  54. [65]

    GLU-Net: Global-local universal network for dense flow and correspondences,

    P. Truong, M. Danelljan, and R. Timofte, “GLU-Net: Global-local universal network for dense flow and correspondences,” inCVPR, 2020

    2020

  55. [66]

    Deep vit features as dense visual descriptors,

    S. Amir, Y . Gandelsman, S. Bagon, and T. Dekel, “Deep vit features as dense visual descriptors,” inECCVW, 2022

    2022

  56. [67]

    Deep semantic feature matching,

    N. Ufer and B. Ommer, “Deep semantic feature matching,” inCVPR, 2017, pp. 5929–5938

    2017

  57. [68]

    Integrative feature and cost aggregation with transformers for dense correspondence,

    S. Hong, S. Cho, S. Kim, and S. Lin, “Integrative feature and cost aggregation with transformers for dense correspondence,”arXiv preprint arXiv:2209.08742, 2022

    2022 arXiv

  58. [69]

    Warp consistency for unsupervised learning of dense correspondences,

    P. Truong, M. Danelljan, F. Yu, and L. Van Gool, “Warp consistency for unsupervised learning of dense correspondences,” inICCV, 2021

    2021

  59. [70]

    Dualrc: A dual-resolution learning framework with neighbourhood consensus for visual corre- spondences,

    X. Li, K. Han, S. Li, and V . Prisacariu, “Dualrc: A dual-resolution learning framework with neighbourhood consensus for visual corre- spondences,”TPAMI, 2024

    2024

  60. [71]

    Dual-resolution correspondence networks,

    ——, “Dual-resolution correspondence networks,” inNeurIPS, 2020

    2020

  61. [72]

    Correspondence transformers with asymmetric feature learning and matching flow super-resolution,

    Y . Sun, D. Zhao, Z. Yin, Y . Huang, T. Gui, W. Zhang, and W. Ge, “Correspondence transformers with asymmetric feature learning and matching flow super-resolution,” inCVPR, 2023

    2023

  62. [73]

    Guided semantic flow,

    S. Jeon, D. Min, S. Kim, J. Choe, and K. Sohn, “Guided semantic flow,” inECCV, 2020

    2020

  63. [74]

    Semi- supervised learning of semantic correspondence with pseudo-labels,

    J. Kim, K. Ryoo, J. Seo, G. Lee, D. Kim, H. Cho, and S. Kim, “Semi- supervised learning of semantic correspondence with pseudo-labels,” inCVPR, 2022

    2022

  64. [75]

    Learning semantic correspondence with sparse annotations,

    S. Huang, L. Yang, B. He, S. Zhang, X. He, and A. Shrivastava, “Learning semantic correspondence with sparse annotations,” inECCV, 2022

    2022

  65. [76]

    Semantic 19 attribute matching networks,

    S. Kim, D. Min, S. Jeong, S. Kim, S. Jeon, and K. Sohn, “Semantic 19 attribute matching networks,” inCVPR, 2019

    2019

  66. [77]

    3×2: 3d object part segmentation by 2d semantic correspondences,

    A. Thai, W. Wang, H. Tang, S. Stojanov, J. M. Rehg, and M. Feiszli, “3×2: 3d object part segmentation by 2d semantic correspondences,” inECCV, 2024

    2024

  67. [78]

    Convolutional hough matching networks for robust and efficient visual correspondence,

    J. Min, S. Kim, and M. Cho, “Convolutional hough matching networks for robust and efficient visual correspondence,”TPAMI, 2023

    2023

  68. [79]

    End-to-end weakly-supervised semantic alignment,

    I. Rocco, R. Arandjelovi ´c, and J. Sivic, “End-to-end weakly-supervised semantic alignment,” inCVPR, 2018

    2018

  69. [80]

    Dctm: Discrete-continuous transformation matching for semantic flow,

    S. Kim, D. Min, S. Lin, and K. Sohn, “Dctm: Discrete-continuous transformation matching for semantic flow,” inICCV, 2017

    2017

  70. [81]

    Gms: Grid-based motion statistics for fast, ultra-robust feature correspondence,

    J. Bian, W.-Y . Lin, Y . Matsushita, S.-K. Yeung, T.-D. Nguyen, and M.-M. Cheng, “Gms: Grid-based motion statistics for fast, ultra-robust feature correspondence,” inCVPR, 2017

    2017

  71. [82]

    Improving ransac’s efficiency with a spatial consistency filter [c],

    T. Sattler, B. Leibe, and L. Kobbelt, “Improving ransac’s efficiency with a spatial consistency filter [c],” inICCV, 2009

    2009

  72. [83]

    Automated scene matching in movies,

    F. Schaffalitzky and A. Zisserman, “Automated scene matching in movies,” inImage and Video Retrieval: International Conference, 2002

    2002

  73. [84]

    Video google: A text retrieval approach to object matching in videos,

    Sivic and Zisserman, “Video google: A text retrieval approach to object matching in videos,” inICCV, 2003

    2003

  74. [85]

    Patchmatch stereo-stereo matching with slanted support windows,

    M. Bleyer, C. Rhemann, and C. Rother, “Patchmatch stereo-stereo matching with slanted support windows,” inBMVC, vol. 11, 2011

    2011

  75. [86]

    Attention is all you need,

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention is all you need,”arXiv preprint arXiv:1706.03762, 2017

    2017 arXiv

  76. [87]

    An image is worth 16x16 words: Transformers for image recognition at scale,

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gelly, J. Uszkoreit, and N. Houlsby, “An image is worth 16x16 words: Transformers for image recognition at scale,” inICLR, 2021

    2021

  77. [88]

    End-to-end object detection with transformers,

    N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, and S. Zagoruyko, “End-to-end object detection with transformers,” in ECCV, 2020

    2020

  78. [89]

    Deformable detr: Deformable transformers for end-to-end object detection,

    X. Zhu, W. Su, L. Lu, B. Li, X. Wang, and J. Dai, “Deformable detr: Deformable transformers for end-to-end object detection,”arXiv preprint arXiv:2010.04159, 2020

    2010 arXiv

  79. [90]

    Loftr: Detector-free local feature matching with transformers,

    J. Sun, Z. Shen, Y . Wang, H. Bao, and X. Zhou, “Loftr: Detector-free local feature matching with transformers,” inCVPR, 2021

    2021

  80. [91]

    Cost aggregation with 4D convolutional swin transformer for few-shot segmentation,

    S. Hong, S. Cho, J. Nam, S. Lin, and S. Kim, “Cost aggregation with 4D convolutional swin transformer for few-shot segmentation,” inECCV, 2022

    2022

  81. [92]

    Swin transformer: Hierarchical vision transformer using shifted windows,

    Z. Liu, Y . Lin, Y . Cao, H. Hu, Y . Wei, Z. Zhang, S. Lin, and B. Guo, “Swin transformer: Hierarchical vision transformer using shifted windows,” inICCV, 2021

    2021

  82. [93]

    Neural matching fields: Implicit representation of matching fields for visual correspondence,

    S. Hong, J. Nam, S. Cho, S. Hong, S. Jeon, D. Min, and S. Kim, “Neural matching fields: Implicit representation of matching fields for visual correspondence,”NeurIPS, 2022

    2022

  83. [94]

    Convolutional neural network architecture for geometric matching,

    I. Rocco, R. Arandjelovic, and J. Sivic, “Convolutional neural network architecture for geometric matching,” inCVPR, 2017

    2017

  84. [95]

    Attentive semantic alignment with offset-aware correlation kernels,

    P. H. Seo, J. Lee, D. Jung, B. Han, and M. Cho, “Attentive semantic alignment with offset-aware correlation kernels,” inECCV, 2018

    2018

  85. [96]

    Recurrent transformer networks for semantic correspondence,

    S. Kim, S. Lin, S. Jeon, D. Min, and K. Sohn, “Recurrent transformer networks for semantic correspondence,” inNeurIPS, 2018

    2018

  86. [97]

    Parn: Pyramidal affine regression networks for dense semantic correspondence,

    S. Jeon, S. Kim, D. Min, and K. Sohn, “Parn: Pyramidal affine regression networks for dense semantic correspondence,” inECCV, 2018

    2018

  87. [98]

    Efficient neighbourhood consensus networks via submanifold sparse convolutions,

    I. Rocco, R. Arandjelovi ´c, and J. Sivic, “Efficient neighbourhood consensus networks via submanifold sparse convolutions,” inECCV, 2020

    2020

  88. [99]

    Dkm: Dense kernelized feature matching for geometry estimation,

    J. Edstedt, I. Athanasiadis, M. Wadenb ¨ack, and M. Felsberg, “Dkm: Dense kernelized feature matching for geometry estimation,” inCVPR, 2023

    2023

  89. [100]

    Weakly supervised learning of semantic correspondence through cas- caded online correspondence refinement,

    Y . Huang, Y . Sun, C. Lai, Q. Xu, X. Wang, X. Shen, and W. Ge, “Weakly supervised learning of semantic correspondence through cas- caded online correspondence refinement,” inICCV, 2023

    2023

  90. [101]

    Imagenet: A large-scale hierarchical image database,

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, “Imagenet: A large-scale hierarchical image database,” inCVPR, 2009

    2009

  91. [102]

    FlowWeb: Joint image set alignment by weaving consistent, pixel-wise correspondences,

    T. Zhou, Y . Jae Lee, S. X. Yu, and A. A. Efros, “FlowWeb: Joint image set alignment by weaving consistent, pixel-wise correspondences,” in CVPR, 2015

    2015

  92. [103]

    Semantic matching by weakly supervised 2d point set registration,

    Z. Laskar, H. R. Tavakoli, and J. Kannala, “Semantic matching by weakly supervised 2d point set registration,”arXiv preprint arXiv:1901.08341, 2019

    1901 arXiv

  93. [104]

    Deep semantic matching with foreground detection and cycle-consistency,

    Y .-C. Chen, P.-H. Huang, L.-Y . Yu, J.-B. Huang, M.-H. Yang, and Y .-Y . Lin, “Deep semantic matching with foreground detection and cycle-consistency,” inACCV, 2018

    2018

  94. [105]

    Learning universal semantic correspondences with no supervision and automatic data curation,

    A. Shtedritski, A. Vedaldi, and C. Rupprecht, “Learning universal semantic correspondences with no supervision and automatic data curation,” 2023

    2023

  95. [106]

    Probabilistic warp consistency for weakly-supervised semantic correspondences,

    P. Truong, M. Danelljan, F. Yu, and L. Van Gool, “Probabilistic warp consistency for weakly-supervised semantic correspondences,” in CVPR, 2022

    2022

  96. [107]

    Learning dense correspondence via 3d-guided cycle consistency,

    T. Zhou, P. Krahenbuhl, M. Aubry, Q. Huang, and A. A. Efros, “Learning dense correspondence via 3d-guided cycle consistency,” in CVPR, 2016

    2016

  97. [108]

    Unpaired image- to-image translation using cycle-consistent adversarial networks,

    J.-Y . Zhu, T. Park, P. Isola, and A. A. Efros, “Unpaired image- to-image translation using cycle-consistent adversarial networks,” in ECCV, 2017

    2017

  98. [109]

    Probabilistic model distillation for semantic correspondence,

    X. Li, D.-P. Fan, F. Yang, A. Luo, H. Cheng, and Z. Liu, “Probabilistic model distillation for semantic correspondence,” inCVPR, 2021

    2021

  99. [110]

    Co-teaching: Robust training of deep neural networks with extremely noisy labels,

    B. Han, Q. Yao, X. Yu, G. Niu, M. Xu, W. Hu, I. Tsang, and M. Sugiyama, “Co-teaching: Robust training of deep neural networks with extremely noisy labels,”NIPS, 2018

    2018

  100. [111]

    Match me if you can: Semantic correspondence learning with unpaired images,

    J. Kim, B. Heo, S. Yun, S. Kim, and D. Han, “Match me if you can: Semantic correspondence learning with unpaired images,”arXiv preprint arXiv:2311.18540, 2023

    2023 arXiv

  101. [112]

    Learning generative visual models from few training examples: An incremental bayesian approach tested on 101 object categories,

    L. Fei-Fei, R. Fergus, and P. Perona, “Learning generative visual models from few training examples: An incremental bayesian approach tested on 101 object categories,” inCVPRW, 2004

    2004

  102. [113]

    One-shot learning of object categories,

    F.-F. Li, R. Fergus, and P. Perona, “One-shot learning of object categories,”TPAMI, 2006

    2006

  103. [114]

    Detect what you can: Detecting and representing objects using holistic models and body parts,

    X. Chen, R. Mottaghi, X. Liu, S. Fidler, R. Urtasun, and A. Yuille, “Detect what you can: Detecting and representing objects using holistic models and body parts,” inCVPR, 2014

    2014

  104. [115]

    Joint recovery of dense corre- spondence and cosegmentation in two images,

    T. Taniai, S. N. Sinha, and Y . Sato, “Joint recovery of dense corre- spondence and cosegmentation in two images,” inCVPR, 2016

    2016

  105. [116]

    Jointly optimizing 3D model fitting and fine-grained classification,

    Y .-L. Lin, V . I. Morariu, W. Hsu, and L. S. Davis, “Jointly optimizing 3D model fitting and fine-grained classification,” inECCV, 2014

    2014

  106. [117]

    Unsupervised joint object discovery and segmentation in internet images,

    M. Rubinstein, A. Joulin, J. Kopf, and C. Liu, “Unsupervised joint object discovery and segmentation in internet images,” inCVPR, 2013

    2013

  107. [118]

    Semantic contours from inverse detectors,

    B. Hariharan, P. Arbel ´aez, L. Bourdev, S. Maji, and J. Malik, “Semantic contours from inverse detectors,” inICCV, 2011

    2011

  108. [119]

    The pascal visual object classes challenge: A retrospective,

    M. Everingham, S. M. A. Eslami, L. Van Gool, C. K. I. Williams, J. Winn, and A. Zisserman, “The pascal visual object classes challenge: A retrospective,”IJCV, 2015

    2015

  109. [120]

    Caltech-256 object category dataset,

    G. Griffin, A. Holub, and P. Perona, “Caltech-256 object category dataset,”CalTech Report, 2007

    2007

  110. [121]

    Learning graphs to match,

    M. Cho, K. Alahari, and J. Ponce, “Learning graphs to match,” in ICCV, 2013

    2013

  111. [122]

    SPair-71k: A large-scale bench- mark for semantic correspondence,

    J. Min, J. Lee, J. Ponce, and M. Cho, “SPair-71k: A large-scale bench- mark for semantic correspondence,”arXiv prepreint arXiv:1908.10543, 2019

    1908 arXiv

  112. [123]

    Beyond pascal: A benchmark for 3d object detection in the wild,

    Y . Xiang, R. Mottaghi, and S. Savarese, “Beyond pascal: A benchmark for 3d object detection in the wild,”WACV, 2014

    2014

  113. [124]

    Ap-10k: A benchmark for animal pose estimation in the wild,

    H. Yu, Y . Xu, J. Zhang, W. Zhao, Z. Guan, and D. Tao, “Ap-10k: A benchmark for animal pose estimation in the wild,” inNeurIPS, 2021

    2021

  114. [125]

    Zero-shot learn- ing—a comprehensive evaluation of the good, the bad and the ugly,

    Y . Xian, C. H. Lampert, B. Schiele, and Z. Akata, “Zero-shot learn- ing—a comprehensive evaluation of the good, the bad and the ugly,” TPAMI, 2018

    2018

  115. [126]

    African wildlife,

    B. Ferreira, “African wildlife,” https://www.kaggle.com/biancaferreira/ african-wildlife, 2021

    2021

  116. [127]

    wild cats,

    E. ˇSahovi´c, “wild cats,” https://www.kaggle.com/enisahovi/ cats-projekat-4, 2020

    2020

  117. [128]

    Animals 5,

    Y . V . Trivedi, “Animals 5,” https://www.kaggle.com/ytrivedi1/ animals-5, 2020

    2020

  118. [129]

    Animal image dataset(dog, cat and panda),

    A. Saxena, “Animal image dataset(dog, cat and panda),” https://www.kaggle.com/ashishsaxena2209/ animal-image-datasetdog-cat-and-panda, 2019

    2019

  119. [130]

    Animals 10,

    C. Alessio, “Animals 10,” https://www.kaggle.com/alessiocorrado99/ animals10, 2019

    2019

  120. [131]

    Iucn animals dataset,

    antoreepjana, “Iucn animals dataset,” https://www.kaggle.com/ antoreepjana/iucn-animals-dataset, 2021

    2021

  121. [132]

    Endangered animals,

    S. Jamil, “Endangered animals,” https://www.kaggle.com/sonain/ endangered-animals, 2020

    2020

  122. [133]

    Misc210k: A large-scale dataset for multi-instance seman- tic correspondence,

    Y . Sun, Y . Huang, H. Guo, Y . Zhao, R. Wu, Y . Yu, W. Ge, and W. Zhang, “Misc210k: A large-scale dataset for multi-instance seman- tic correspondence,” inCVPR, 2023

    2023

  123. [134]

    Semantic correspondence with transformers,

    S. Cho, S. Hong, S. Jeon, Y . Lee, K. Sohn, and S. Kim, “Semantic correspondence with transformers,”arXiv preprint arXiv:2106.02520, 2021

    2021 arXiv

  124. [135]

    Emerging properties in self-supervised vision transformers,

    M. Caron, H. Touvron, I. Misra, H. J ´egou, J. Mairal, P. Bojanowski, and A. Joulin, “Emerging properties in self-supervised vision transformers,” inICCV, 2021

    2021

  125. [136]

    Demystifying unsupervised semantic correspondence estimation,

    M. Ayg ¨un and O. Mac Aodha, “Demystifying unsupervised semantic correspondence estimation,” inECCV, 2022

    2022

  126. [137]

    Asic: Aligning sparse in-the-wild image collections,

    K. Gupta, V . Jampani, C. Esteves, A. Shrivastava, A. Makadia, 20 N. Snavely, and A. Kar, “Asic: Aligning sparse in-the-wild image collections,” inICCV, 2023

    2023

  127. [138]

    Semantic correspondence as an optimal transport problem,

    Y . Liu, L. Zhu, M. Yamada, and Y . Yang, “Semantic correspondence as an optimal transport problem,” inCVPR, 2020

    2020

  128. [139]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inCVPR, 2016

    2016

  129. [140]

    Learning transferable visual models from natural language supervision,

    A. Radford, J. W. Kim, C. Hallacy, A. Ramesh, G. Goh, S. Agarwal, G. Sastry, A. Askell, P. Mishkin, J. Clarket al., “Learning transferable visual models from natural language supervision,” inICML, 2021

    2021

  130. [141]

    ibot: Image bert pre-training with online tokenizer,

    J. Zhou, C. Wei, H. Wang, W. Shen, C. Xie, A. Yuille, and T. Kong, “ibot: Image bert pre-training with online tokenizer,”ICLR, 2022. 1 Overview.In this appendix, we present detailed exper- imental evaluations and qualitative analysis to support our main findings. In Section A,...

    2022

Pith tools

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