Pith. sign in

REVIEW 5 major objections 6 minor 3 cited by

Multispecies Animal Re-ID Using a Large Community-Curated Dataset

T0 review · 5 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read One AI model trained on 49 species beats separate single-species models, with a 12.5% average top-1 accuracy gain.

desk verdict A solid empirical study on multi-species animal re-id: the dataset and experiments are valuable, but the reported gains need error bars and a defense against label-bias before I'd trust the exact numbers. read the letter →

arxiv 2412.05602 v1 pith:Z6DABLX2 submitted 2024-12-07 cs.CV

Lasha Otarashvili , Tamilselvan Subramanian , Jason Holmberg , J.J. Levenson , Charles V. Stewart This is my paper · ORCID
classification cs.CV
keywords animalre-identificationmulti-speciestrainingmetriclearningcommunity-curateddatasetzero-shottransferfine-tuningwildlifemonitoringconservationtechnology
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper argues that the usual practice of training a separate re-identification model for each animal species is both costly and wasteful, and that a single model trained jointly on many species can do better. To test this, the authors assembled a community-curated dataset of 49 species, 37,138 individual animals, and 225,374 annotated images, and trained one embedding network on it. They report that this multi-species model beats the same architecture trained species-by-species by an average of 12.5% in top-1 accuracy, with the largest gains on species that have the least data. If true, this means a wildlife monitoring system can replace many per-species systems with one model, and can bootstrap identification for new species from very few labelled images.

What carries the argument

The machinery is a shared embedding space learned by a single EfficientNetV2-M backbone with a sub-center ArcFace loss (k=3) that uses dynamic margins scaled by class frequency. The classification layer is discarded at inference and each image is mapped to a descriptor; matching is done by cosine distance. The argument is that training on many species forces the network to encode generic 'individual identity' features, such as spot patterns, fin shapes, and body markings, that transfer across species, which is why the multi-species model holds up for rare species and for species never seen in training.

What would settle it

Re-run the same multi-species vs. single-species comparison on a dataset where individual identities are confirmed independently, for example by genetic sampling or by expert photo-id done without any algorithmic ranking; if the 12.5% top-1 advantage shrinks or disappears, the gain is largely an artifact of how the training labels were curated. A second check is to compare the identity labels between datasets for the same species, such as the turtle sets with and without the '+head' suffix, and verify that the same individual animals carry consistent identities across both; inconsistent labels would inflate the reported accuracy.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that joint training across species is not a compromise but an advantage: a single MiewID model trained on 49 species consistently outperforms identical models trained separately on each species, with an average top-1 accuracy gain of 12.5% and per-species gains ranging from -0.2% to +77.3%. The same model, when evaluated on species it never saw during training, outperforms the MegaDescriptor baseline on all 33 species tested, with an average top-1 improvement of 19.2%. The authors also show that adding just a small number of annotations for a new species to the multi-species training set, or fine-tuning the published model, produces better accuracy than training that species from scratch, and that the model is already used in production for more than 60 species.

Load-bearing premise

The ground-truth identities were produced by community curation that sometimes relied on earlier computer-vision algorithms to rank candidate matches, so the labels may contain patterns that make the measured accuracy look better than it is in the wild.

Editorial extensions

If this is right

  • Wildlife monitoring platforms can deploy one model instead of dozens of per-species models, cutting training, storage, and maintenance costs.
  • Species with very little labeled data benefit the most from joint training, reducing the annotation burden for rare and endangered animals.
  • New species can be added by fine-tuning the released model or by adding a few hundred annotations to the training set, rather than building a dataset and model from scratch.
  • Zero-shot matching on unseen species is strong enough to help human curators rank candidate identities, which is the first step in bootstrapping a new species in a monitoring system.

Reading between the lines

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

  • If the gain comes from shared visual structure across species, then deliberately increasing taxonomic diversity in the training set may matter more than simply adding more images of well-sampled species, a prediction the paper's data supports but does not directly test.
  • Because the curation labels were themselves guided by earlier algorithms such as Hotspotter and integral curvature matching, the reported boost could partly reflect those algorithms' biases; an independent genetic or fully manual ground truth would be needed to separate the model's true advantage from label bias.
  • The same architecture's success across such different body plans suggests that a single frozen embedding, with small per-species adaptations, could grow into a general animal-identification foundation model that conservation groups share rather than each building their own.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 6 minor

Summary. The paper introduces MiewID, a multi-species animal re-identification model trained on a community-curated dataset of 49 species, 37K individuals, and 225K images. The central claims are (i) a single multi-species model consistently outperforms per-species models by 12.5% average top-1 accuracy (Section 5.1, Table 2); (ii) in zero-shot transfer to 33 unseen species it outperforms MegaDescriptor on every species, with an average top-1 gain of 19.2% (Section 5.3, Table 3); and (iii) the model is effective for bootstrapping and fine-tuning new species with limited data (Sections 5.4 and 5.5). The paper also releases code and model weights. The experiments are extensive, with leave-one-out analyses and comparisons across backbones, but the evaluation protocol and label-construction process raise concerns about the strength of the central quantitative claims.

Significance. If the central claims hold, multi-species training is not merely a practical convenience but a genuine performance advantage over per-species models, and the demonstrated zero-shot gains would be an important step toward scalable wildlife re-identification. The paper's main strengths are the scale and diversity of the assembled dataset, the transparency about the community-curation process, the direct comparison against a recent strong baseline (MegaDescriptor), and the public release of code and weights, which should facilitate reproducibility and follow-up work. The leave-one-out and few-shot experiments are thoughtful and address a real deployment need. However, the absence of uncertainty estimates, the mixed seen/unseen test set, and the lack of a clean-label validation set mean the headline numbers should be treated with caution until these gaps are addressed.

major comments (5)
  1. [Section 3.3, Table 2] The test set deliberately mixes 50% individuals seen during training (with different images) and 50% unseen individuals, but the reported metrics are averaged over this mixed set without a separate breakdown. Because closed-set matches to known individuals are generally easier than open-set identification, the claimed 12.5% average gain over single-species models may be inflated by the seen-individual portion. The authors should report top-1 accuracy separately for seen and unseen individuals, and confirm that the multi-species advantage persists on the unseen-only subset, which is the regime that motivates the paper.
  2. [Section 3.1, Section 5.3] The ground-truth identities were assigned through a curation process in which 'users employ manual photo ID ... and earlier re-identification computer vision algorithms [5,20,34] to rank-order potential individual matches.' This means the labels may encode the biases of the very algorithms the model is compared against, particularly in the zero-shot comparison with MegaDescriptor: MiewID was trained on species curated with the same Wildbook pipeline, while MegaDescriptor was trained on a different distribution. The 19.2% average zero-shot improvement could therefore partly reflect familiarity with curation-specific visual cues rather than general re-identification ability. The paper should validate the model on an independently curated, algorithm-free label set, or at least quantify the degree of label automation per species and show that the conclusions are robust when restricted to manually curated subsets.
  3. [Section 5.1, Section 5.3] All reported accuracies are point estimates from a single training run, with no error bars, confidence intervals, or repeated-seed results. The claim that the multi-species model 'consistently outperforms' single-species models is supported only by per-species point estimates whose variance is unknown; a single negative species (-0.2%) and per-species gains ranging up to 77.3% suggest high variance. Similarly, the 'average top-1 improvement of 19.2%' over MegaDescriptor is a single-run comparison. The authors should provide repeated-seed statistics (at least for the main comparisons) or otherwise justify that the differences are not within run-to-run noise.
  4. [Section 3.4] The evaluation protocol uses the test set as both gallery and query in a one-vs-all scheme, with a restriction of at most one annotation per encounter. While the authors justify this as avoiding the 'soft data leak' of using the training set as gallery, the resulting protocol is non-standard and may not be directly comparable to results in the literature. For instance, the gallery and query are drawn from the same set of images, which can overestimate retrieval accuracy compared to a fixed, independent gallery. The paper should compare its protocol against the standard train-as-gallery/test-as-query protocol on at least a subset of species, or provide a clear argument for why the one-vs-all protocol yields accurate estimates for real deployment.
  5. [Section 5.3, Table 3] The comparison with MegaDescriptor is confounded by resolution (MiewID at 256px vs MegaDescriptor-L-384 at 384px) and by architecture/backbone differences. The paper states both are evaluated at their native resolutions, but this makes the comparison one of entire systems rather than a controlled test of multi-species training. The 19.2% average improvement could be due to higher input resolution, different backbone capacity, or different training data, rather than the multi-species training approach itself. The authors should either run a controlled comparison at matched input resolution and backbone, or explicitly characterize which factors drive the gain.
minor comments (6)
  1. [Abstract] The phrase 'averaging an 19.2% top-1 improvement' contains a grammatical error ('an' should be 'a'); the same issue appears elsewhere in the paper.
  2. [Section 4] The first sentence reads 'Our model model starts from an EfficientNetV2-M [29] backbone'; the duplicated word 'model' should be removed.
  3. [Section 3.1] There is a typo in 'a set of 0 or more "annnotations" per image'; 'annnotations' should be 'annotations'.
  4. [Table 4] The formatting of Table 4 is badly garbled in the provided text, with rows and columns merged incorrectly; the table should be regenerated in a clean, readable format.
  5. [References] References [20] and [21] both refer to the same work by Moskvyak et al. on manta ray re-identification; one entry should be removed or the duplicates distinguished.
  6. [Section 5.2] The phrase 'no annotations from the species are seen during' appears to be missing the word 'training' at the end of the sentence.

Circularity Check

1 steps flagged · score 2.0 of 10

The benchmark is largely self-contained, but the community-curated ground-truth labels were partly generated with the help of earlier re-ID algorithms from the same research lineage, so the headline gains may partly reflect agreement with those algorithms' suggestions.

  1. other [Section 3, Data, first paragraph]
    "Inside Wildbook, users employ manual photo ID (i.e., “by eye”) and earlier re-identification computer vision algorithms [5,20,34] to rank-order potential individual matches and then make curation decisions about individual animal identities across sightings directly in the software."

    The ground-truth identities used to compute all reported top-1 accuracies (Tables 2 and 3) were produced by a curation process that included earlier re-identification algorithms [5,20,34], several of which are from the same research lineage as the present model (e.g., Hotspotter [5] and Integral Curvature [34], with overlapping authors). The model is then scored by how often its top-1 match agrees with these algorithm-assisted labels. Thus the headline 12.5% and 19.2% gains partly measure the model's ability to reproduce the same visual matching cues that helped create the labels, rather than purely independent biological identity.

full rationale

This is an empirical benchmark paper rather than a derivation, and most of the reported comparisons are self-contained: the multi-species versus single-species models are trained and tested on the same splits, and the zero-shot comparison against MegaDescriptor is an external, independent benchmark. No fitted parameter is renamed as a prediction, and no uniqueness theorem is imported from the authors' prior work. The only circularity-adjacent issue is the provenance of the ground-truth identities: Section 3 states that Wildbook curators used earlier re-identification computer vision algorithms, several from the same research group, to rank-order candidate matches before making final identity decisions. Because these algorithm-assisted labels are the target of every top-1 accuracy measurement, the headline gains could partly reflect agreement with the earlier algorithms' matching cues rather than independent biological identity. The paper is transparent that final decisions were made by human experts, so this is an indirect, partial circularity rather than a full reduction; it does not invalidate the empirical comparison but means the absolute accuracy numbers should be confirmed on a clean-label validation set. The absence of a separate validation set and the 50% seen-individual test composition are additional limitations but are not circular.

Assumptions & free parameters 5 free parameters · 3 assumptions · 0 invented entities

The central claims rest on the reliability of community-curated identity labels, the suitability of the view-based identity definition, and the evaluation protocol. No new physical or conceptual entities are introduced.

free parameters (5)
  • ArcFace scale = 51.5
    A fixed constant in the sub-center ArcFace loss, chosen by the authors without a stated justification.
  • Dynamic margin initialization = 0.5
    The margin starts at 0.5 and is adjusted by class frequency, effectively tuned to the training set distribution.
  • Sub-center count k = 3
    Number of sub-centers in ArcFace, selected as a hyperparameter.
  • Training image size for experiments = 256x256
    Used for the comparison experiments, while the production version uses 440x440.
  • Training hyperparameters (batch size, learning rate, augmentation) = batch 112, warmup 15 epochs, decay 0.8
    Optimized with Optuna on random splits, so these numbers are data-dependent choices.
assumptions (3)
  • domain assumption Ground-truth identity labels from expert curation are correct and consistent across datasets.
    The entire evaluation depends on these labels, but curation sometimes uses algorithmic suggestions (Section 3).
  • domain assumption Treating different viewpoints as different individuals for some species is appropriate for the re-id task.
    For species like cheetahs, left and right sides are so different that matching them is impractical; the paper splits them into separate IDs (Section 3.2).
  • domain assumption The test set construction (max 10 samples per individual, one per encounter) yields a representative and unbiased evaluation.
    This filtering removes near-duplicate images, but it also changes the difficulty of the retrieval task (Section 3.3).

how reviews work

0 comments
Cite this review

Pith. "Pith review of Multispecies Animal Re-ID Using a Large Community-Curated Dataset." pith.science (2026). https://pith.science/paper/Z6DABLX2

@misc{pith2026241205602,
  author       = {Pith},
  title        = {Pith review of: Multispecies Animal Re-ID Using a Large Community-Curated Dataset},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Z6DABLX2}},
  note         = {Machine review of arXiv:2412.05602}
}
read the original abstract

Recent work has established the ecological importance of developing algorithms for identifying animals individually from images. Typically, a separate algorithm is trained for each species, a natural step but one that creates significant barriers to wide-spread use: (1) each effort is expensive, requiring data collection, data curation, and model training, deployment, and maintenance, (2) there is little training data for many species, and (3) commonalities in appearance across species are not exploited. We propose an alternative approach focused on training multi-species individual identification (re-id) models. We construct a dataset that includes 49 species, 37K individual animals, and 225K images, using this data to train a single embedding network for all species. Our model employs an EfficientNetV2 backbone and a sub-center ArcFace loss function with dynamic margins. We evaluate the performance of this multispecies model in several ways. Most notably, we demonstrate that it consistently outperforms models trained separately on each species, achieving an average gain of 12.5% in top-1 accuracy. Furthermore, the model demonstrates strong zero-shot performance and fine-tuning capabilities for new species with limited training data, enabling effective curation of new species through both incremental addition of data to the training set and fine-tuning without the original data. Additionally, our model surpasses the recent MegaDescriptor on unseen species, averaging an 19.2% top-1 improvement per species and showing gains across all 33 species tested. The fully-featured code repository is publicly available on GitHub, and the feature extractor model can be accessed on HuggingFace for seamless integration with wildlife re-identification pipelines. The model is already in production use for 60+ species in a large-scale wildlife monitoring system.

Figures

Figures reproduced from arXiv: 2412.05602 by the authors.

Figure 1
Figure 1. Comparison of top-1 performance of multi-species and single-species models. Species are ordered by increasing numbers of [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Top-5 performance comparison per species for model trained on the full dataset and model trained on all species except the test [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Mean top-1 accuracy and standard deviation across [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Comparison of Swin-V2 and Efficientnet-V2 backbones [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Comparison of top-1 performance of multi-species and single-species models. Species are ordered by increasing numbers of [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: Comparison of top-5 performance of multi-species and single-species models. Species are ordered by increasing numbers of [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Comparison of top-10 performance of multi-species and single-species models. Species are ordered by increasing numbers of [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]
Figure 8
Figure 8. Figure 8: Comparison of top-20 performance of multi-species and single-species models. Species are ordered by increasing numbers of [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: Top-1 performance comparison per species for model trained on the full dataset and model trained on all species [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 10
Figure 10. Figure 10: Top-5 performance comparison per species for model trained on the full dataset and model trained on all species [PITH_FULL_IMAGE:figures/full_fig_p013_10.png]
Figure 11
Figure 11. Figure 11: Top-10 performance comparison per species for model trained on the full dataset and model trained on all species [PITH_FULL_IMAGE:figures/full_fig_p013_11.png]
Figure 12
Figure 12. Figure 12: Top-20 performance comparison per species for model trained on the full dataset and model trained on all species [PITH_FULL_IMAGE:figures/full_fig_p013_12.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 3 Pith papers

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

  1. DS@GT ARC at AnimalCLEF 2026: Species-Aware Graph Construction for Multi-Species Animal Re-Identification

    cs.CV 2026-07 conditional novelty 6.0 of 10

    A species-aware graph-construction pipeline with global retrieval, LightGlue matching, LightGBM scoring, and Leiden clustering reached private ARI 0.674 and 5th place in AnimalCLEF 2026.

  2. Wildlife Target Re-Identification Using Self-supervised Learning in Non-Urban Settings

    cs.CV 2025-07 reject novelty 5.0 of 10

    Self-supervised wildlife re-identification using temporal camera trap pairs is claimed to outperform supervised methods, but the experiments do not control for training data and hence do not support the claim as stated.

  3. Calibrated Similarity and Graph Clustering for Open-Set Animal Re-Identification

    cs.CV 2026-08 conditional novelty 4.0 of 10

    A calibrated fusion pipeline with segmentation, species-specific preprocessing, and graph clustering reached top public (0.721) and private (0.711) ARI scores on the AnimalCLEF26 open-set animal re-identification benchmark.

Reference graph

Works this paper leans on

38 extracted references · 36 canonical work pages · cited by 3 Pith papers

  1. [1]

    Seaturtleid2022: A long-span dataset for re- liable sea turtle re-identification

    Luk ´aˇs Adam, V ojt ˇech ˇCerm´ak, Kostas Papafitsoros, and Lukas Picek. Seaturtleid2022: A long-span dataset for re- liable sea turtle re-identification. In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV), pages 7146–7156, January 2024. 2, 4

    work page 2024

  2. [2]

    Berger-Wolf, Daniel I

    Tanya Y . Berger-Wolf, Daniel I. Rubenstein, Charles V . Stewart, Jason A. Holmberg, Jason Parham, Shyamala Menon, Josh Crall, John Van Oast, Emre Kiciman, and Lucas Joppa. Wildbook: Crowdsourcing, computer vision, and data science for conservation. arXiv preprint arXiv:1710.08880,

    arXiv

  3. [3]

    Wildfusion: Individual animal identi- fication with calibrated similarity fusion, 2024

    V ojtˇech Cermak, Lukas Picek, Luk ´aˇs Adam, Luk ´aˇs Neu- mann, and Jiˇr´ı Matas. Wildfusion: Individual animal identi- fication with calibrated similarity fusion, 2024. 2, 4

    work page 2024

  4. [4]

    Advanced image recognition: a fully automated, high- accuracy photo-identification matching system for hump- back whales

    Ted Cheeseman, Ken Southerland, Jinmo Park, Marilia Olio, Kiirsten Flynn, John Calambokidis, Lindsey Jones, Claire Garrigue, Astrid Frisch Jord ´an, Addison Howard, Walter Reade, Janet Neilson, Christine Gabriele, and Phil Clapham. Advanced image recognition: a fully automated, high- accuracy photo-identification matching system for hump- back whales. Mam...

    work page 2022

  5. [5]

    Crall, Charles V

    James P. Crall, Charles V . Stewart, Tanya Y . Berger-Wolf, Daniel I. Rubenstein, and Siva R. Sundaresan. Hotspotter - patterned species instance recognition. In Proceedings of the IEEE Winter Conference on Applications of Computer Vision (WACV), 2013. 2

    work page 2013

  6. [6]

    Sub-center arcface: Boosting face recognition by large-scale noisy web faces

    Jiankang Deng, Jia Guo, Tongliang Liu, Mingming Gong, and Stefanos Zafeiriou. Sub-center arcface: Boosting face recognition by large-scale noisy web faces. In Proceedings of the European Conference on Computer Vision (ECCV) , pages 741–757. Springer, 2020. 4

    work page 2020

  7. [7]

    Arcface: Additive angular mar- gin loss for deep face recognition.IEEE Transactions on Pat- tern Analysis and Machine Intelligence , 44(10):5962–5979, Oct

    Jiankang Deng, Jia Guo, Jing Yang, Niannan Xue, Irene Kot- sia, and Stefanos Zafeiriou. Arcface: Additive angular mar- gin loss for deep face recognition.IEEE Transactions on Pat- tern Analysis and Machine Intelligence , 44(10):5962–5979, Oct. 2022. 1 8

    work page 2022

  8. [8]

    Automated identification of individuals in wildlife population using siamese neural networks

    Nkosikhona Dlamini and Terence L van Zyl. Automated identification of individuals in wildlife population using siamese neural networks. In 2020 7th International Con- ference on Soft Computing & Machine Intelligence (ISCMI), pages 224–228. IEEE, 2020. 2

    work page 2020

Show all 38 references
  1. [9]

    Chimpanzee faces in the wild: Log-euclidean cnns for predicting iden- tities and attributes of primates

    Alexander Freytag, Erik Rodner, Marcel Simon, Alexander Loos, Hjalmar K ¨uhl, and Joachim Denzler. Chimpanzee faces in the wild: Log-euclidean cnns for predicting iden- tities and attributes of primates. In German Conference on Pattern Recognition (GCPR), 2016. 2

    2016

  2. [10]

    Google landmark recognition 2020 competition third place solution,

    Qishen Ha, Bo Liu, Fuxu Liu, and Peiyuan Liao. Google landmark recognition 2020 competition third place solution,

    2020

  3. [11]

    Toward re-identifying any animal

    Bingliang Jiao, Lingqiao Liu, Liying Gao, Ruiqi Wu, Gu- osheng Lin, PENG W ANG, and Yanning Zhang. Toward re-identifying any animal. In Thirty-seventh Conference on Neural Information Processing Systems, 2023. 2

    2023

  4. [12]

    Atrw dataset: Ai city challenge track 5 - animal track re-id in the wild, 2022

    Labeled Information Library of Alexandria (LiLA). Atrw dataset: Ai city challenge track 5 - animal track re-id in the wild, 2022. Accessed: 2024-07-02. 2

    2022

  5. [13]

    Beluga id 2022 dataset, 2022

    Labeled Information Library of Alexandria (LiLA). Beluga id 2022 dataset, 2022. Accessed: 2024-07-02. 2

    2022

  6. [14]

    Hyena id 2022 dataset, 2022

    Labeled Information Library of Alexandria (LiLA). Hyena id 2022 dataset, 2022. Accessed: 2024-07-02. 2

    2022

  7. [15]

    Leopard id 2022 dataset, 2022

    Labeled Information Library of Alexandria (LiLA). Leopard id 2022 dataset, 2022. 2024-07-02. 2

    2022

  8. [16]

    Whale shark id dataset, 2022

    Labeled Information Library of Alexandria (LiLA). Whale shark id dataset, 2022. Accessed: 2024-07-02. 2

    2022

  9. [17]

    Labeled information library of alexandria: Biol- ogy and conservation

    LILA BC. Labeled information library of alexandria: Biol- ogy and conservation. 2

  10. [18]

    Lightglue: Local feature matching at light speed, 2023

    Philipp Lindenberger, Paul-Edouard Sarlin, and Marc Polle- feys. Lightglue: Local feature matching at light speed, 2023. 2

    2023

  11. [19]

    Swin transformer v2: Scaling up capacity and resolution

    Ze Liu, Han Hu, Yutong Lin, Zhuliang Yao, Zhenda Xie, Yixuan Wei, Jia Ning, Yue Cao, Zheng Zhang, Li Dong, Furu Wei, and Baining Guo. Swin transformer v2: Scaling up capacity and resolution. In International Conference on Computer Vision and Pattern Recognition (CVPR), 2022. 7

    2022

  12. [20]

    Armstrong, Feras Dayoub, and Mahsa Baktash

    Olga Moskvyak, Fr ´ed´eric Maire, Asia O. Armstrong, Feras Dayoub, and Mahsa Baktash. Robust re-identification of manta rays from natural markings by learning pose invariant embeddings. 2021 Digital Image Computing: Techniques and Applications (DICTA), pages 1–8, 2019. 2

    2021

  13. [21]

    Armstrong, Feras Dayoub, and Mahsa Baktashmotlagh

    Olga Moskvyak, Frederic Maire, Asia O. Armstrong, Feras Dayoub, and Mahsa Baktashmotlagh. Robust re- identification of manta rays from natural markings by learn- ing pose invariant embeddings, 2019. 2

    2019

  14. [23]

    A deep learning approach for dog face verification and recogni- tion

    Guillaume Mougeot, Dewei Li, and Shuai Jia. A deep learning approach for dog face verification and recogni- tion. In Abhaya C. Nayak and Alok Sharma, editors, PRI- CAI 2019: Trends in Artificial Intelligence , pages 418–430, Cham, 2019. Springer International Publishing. 2

    2019

  15. [24]

    Norppa: Novel ringed seal re- identification by pelage pattern aggregation, 2022

    Ekaterina Nepovinnykh, Ilia Chelak, Tuomas Eerola, and Heikki K ¨alvi¨ainen. Norppa: Novel ringed seal re- identification by pelage pattern aggregation, 2022. 2

    2022

  16. [25]

    Sealid: Saimaa ringed seal re-identification dataset

    Ekaterina Nepovinnykh, Tuomas Eerola, Vincent Biard, Piia Mutka, Marja Niemi, Mervi Kunnasranta, and Heikki K¨alvi¨ainen. Sealid: Saimaa ringed seal re-identification dataset. Sensors, 22(19), 2022. 2

    2022

  17. [26]

    Rubenstein

    Jason Parham, Jonathan Crall, Charles Stewart, Tanya Berger-Wolf, and Daniel I. Rubenstein. Animal popula- tion censusing at scale with citizen science and photographic identification. In AAAI Spring Symposium-Technical Report, Palo Alto, CA, January 1 2017. AAAI Press. 2

    2017

  18. [27]

    Berg, and Li Fei-Fei

    Olga Russakovsky, Jia Deng, Hao Su, Jonathan Krause, San- jeev Satheesh, Sean Ma, Zhiheng Huang, Andrej Karpathy, Aditya Khosla, Michael Bernstein, Alexander C. Berg, and Li Fei-Fei. ImageNet Large Scale Visual Recognition Chal- lenge. International Journal of Computer Vision ...

    2015

  19. [28]

    Superglue: Learning feature matching with graph neural networks, 2020

    Paul-Edouard Sarlin, Daniel DeTone, Tomasz Malisiewicz, and Andrew Rabinovich. Superglue: Learning feature matching with graph neural networks, 2020. 2

    2020

  20. [29]

    Efficientnetv2: Smaller models and faster training

    Mingxing Tan and Quoc Le. Efficientnetv2: Smaller models and faster training. In International conference on machine learning, pages 10096–10106. PMLR, 2021. 4

    2021

  21. [30]

    finfindr : Automated recognition and identification of marine mammal dorsal fins using residual convolutional neural networks

    Jaime Thompson, Victoria Zero, Lori Schwacke, Todd Speakman, Brian Quigley, Jeanine Morey, and Trent Mc- donald. finfindr : Automated recognition and identification of marine mammal dorsal fins using residual convolutional neural networks. Marine Mammal Science, 38, 07 2021. 2

    2021

  22. [31]

    Primface dataset, 2024

    Visiome Platform. Primface dataset, 2024. Accessed: 2024- 07-02. The face images used in this study are provided by PrimFace database: http://visiome.neuroinf.jp/primface, funded by Grant-in-Aid for Scientific research on Innova- tive Areas, ”Face Perception and Recognition” f...

    2024

  23. [32]

    A survey of face recognition,

    Xinyi Wang, Jianteng Peng, Sufang Zhang, Bihui Chen, Yi Wang, and Yandong Guo. A survey of face recognition,

  24. [33]

    Paul, Anka Bedetti, Michelle Henley, Jerenimo Lepirei, and Frank Pope

    Hendrik Weideman, Charles Stewart, Jason Parham, Jason Holmberg, Kiirsten Flynn, John Calambokidis, D. Paul, Anka Bedetti, Michelle Henley, Jerenimo Lepirei, and Frank Pope. Extracting identifying contours for african elephants and humpback whales using a learned appearance mo...

    2020

  25. [34]

    Weideman, Zachary M

    Hendrik J. Weideman, Zachary M. Jablons, Jason A. Holm- berg, Kiirsten Flynn, John Calambokidis, Reny B. Tyson, Ja- son B. Allen, Randall S. Wells, Krista Hupman, Kim Urian, and Charles V . Stewart. Integral curvature representation and matching algorithms for identification o...

    2017

  26. [35]

    Beluga ID 2022, 2022

    WildMe. Beluga ID 2022, 2022. 2

    2022

  27. [36]

    Leopard ID 2022, 2022

    WildMe. Leopard ID 2022, 2022. 2

    2022

  28. [37]

    Macaque faces

    Claire Witham and Emily Bethell. Macaque faces. figshare. Dataset, 2019. 2 9

    2019

  29. [38]

    Mang Ye, Jianbing Shen, Gaojie Lin, Tao Xiang, Ling Shao, and Steven C. H. Hoi. Deep learning for person re- identification: A survey and outlook, 2021. 1

    2021

  30. [39]

    Wildlifedatasets: An open-source toolkit for ani- mal re-identification, 2023

    V ojtˇech ˇCerm´ak, Lukas Picek, Luk´aˇs Adam, and Kostas Pa- pafitsoros. Wildlifedatasets: An open-source toolkit for ani- mal re-identification, 2023. 1, 2, 4, 5 10 A. Appendix A.1. Training Optimized with optuna. Augmentation: rand color sharpening, clahe, shift (25%), scal...

    2023

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.