Pith. sign in

REVIEW 2 major objections 2 minor 45 references

Transfer learning from human faces and ImageNet enables accurate face recognition for dogs, primates, and cattle despite small datasets.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.3

2026-06-27 16:54 UTC pith:ODP7YXFS

load-bearing objection Transfer learning from human faces and ImageNet produces usable numbers on animal faces, but the gains track image quality so closely that the benefit of pre-training itself is not isolated. the 2 major comments →

arxiv 2606.09353 v1 pith:ODP7YXFS submitted 2026-06-08 cs.CV cs.AI

Beyond Humans: Multispecies Animal Face Recognition Using Transfer Learning

Maria De Marsico , Anil K. Jain , Annalaura Miglino This is my paper
classification cs.CV cs.AI
keywords animal face recognitiontransfer learningFaceNetVision Transformerdog identificationcattle recognitionprimate face recognitionmultispecies biometrics
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper investigates whether deep networks pre-trained on human faces or general objects can recognize individual animals by their faces when only limited animal images are available. This approach could replace microchips and other physical tags for tracking lost pets, endangered wildlife, and farm animals, offering a non-invasive alternative that works at a distance. Experiments use FaceNet trained on human faces and the Vision Transformer pre-trained on ImageNet, applied to three datasets: dogs with high-quality images, cattle with medium quality, and primates with lower quality and more variation. Results show the best performance on dogs with ViT reaching 96.85 percent verification accuracy, competitive or superior results on cattle, and promising but variable outcomes on primates. The findings indicate that transfer learning can overcome the data scarcity problem in animal biometrics.

Core claim

Models pre-trained on human faces (FaceNet) or ImageNet (ViT) can be transferred to animal face recognition tasks on dogs, primates, and cattle, achieving high verification accuracy and identification rates that often match or exceed those of networks trained ad hoc on each animal dataset.

What carries the argument

Transfer learning with FaceNet (pre-trained on human faces) and Vision Transformer (pre-trained on ImageNet) used as backbone networks to adapt features to small per-subject animal face datasets.

Load-bearing premise

Differences in image quality, pose variation, and species-specific facial features across the datasets do not prevent useful transfer from human-face or ImageNet pre-training.

What would settle it

A test where the same animal datasets are used to train networks from random initialization and those networks match or exceed the accuracy of the transferred FaceNet and ViT models would show that pre-training provides no advantage.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • ViT achieves a mean verification accuracy of 96.85% and Rank-1 identification rate of 84.34% on the dog dataset.
  • ViT results outperform state-of-the-art ad hoc networks on the cattle dataset.
  • FaceNet remains competitive on cattle while ViT leads overall.
  • Results for primates vary by species and task but remain encouraging for verification and identification.
  • Face recognition offers a non-invasive alternative to physical devices for individual animal tracking in pets, conservation, and farming.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same transfer strategy might apply to additional mammal species whose facial structures share enough common features with the tested ones.
  • Real-world use would benefit from further tests on datasets that include greater motion blur or extreme poses than those already studied.
  • Combining outputs from multiple pre-trained sources could reduce performance variation seen on the primate datasets.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 2 minor

Summary. The paper investigates transfer learning for multispecies animal face recognition using FaceNet (pre-trained on human faces) and ViT (pre-trained on ImageNet) as backbones on three datasets: dogs, primates (lemurs, golden monkeys, chimpanzees), and cattle. It reports that performance tracks image quality (highest on dogs with ViT at 96.85% mean verification accuracy and 84.34% Rank-1 identification), that ViT outperforms SOTA on cattle while FaceNet remains competitive, and that results on primates are encouraging but mixed and do not always beat ad-hoc SOTA models. The central claim is that pre-trained human/object models enable useful recognition without large animal-specific training sets.

Significance. If the empirical results hold after addressing controls, the work would demonstrate practical transfer from human-face and ImageNet pre-training to animal faces across species with varying data quality, supporting non-invasive identification applications in conservation, pet recovery, and livestock management. The explicit comparison to SOTA on multiple tasks (verification and identification) and the ordering of datasets by quality provide a useful baseline, though the absence of ablations limits claims about the source of the gains.

major comments (2)
  1. [Abstract / Results] Abstract and Results section: the reported mean accuracies (e.g., ViT dogs verification 96.85%, Rank-1 84.34%) and SOTA comparisons lack error bars, number of runs, dataset sizes per subject, or statistical tests, so it is impossible to assess whether outperformance on cattle or the ordering across species is reliable rather than noise.
  2. [Experiments / Discussion] Experiments / Discussion: the manuscript states that image quality (resolution, blur, pose) decreases from dogs to cattle to primates and that performance tracks this ordering, yet provides no ablation (resolution-matched subsets, from-scratch training on augmented animal data, or quality-controlled baselines) to isolate the contribution of FaceNet/ViT pre-training from dataset difficulty; this directly affects the transfer-learning claim.
minor comments (2)
  1. [Methods] Clarify the exact train/test splits, number of identities, and images per identity for each dataset to allow reproducibility of the reported figures.
  2. [Methods] Specify the fine-tuning protocol (learning rate, epochs, loss, augmentation) for both FaceNet and ViT so readers can judge how much adaptation occurred versus pure transfer.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript investigating transfer learning for multispecies animal face recognition. The comments highlight important aspects of statistical reporting and experimental controls that we will address to strengthen the work. Below we respond point by point to the major comments.

read point-by-point responses

  1. Referee: [Abstract / Results] Abstract and Results section: the reported mean accuracies (e.g., ViT dogs verification 96.85%, Rank-1 84.34%) and SOTA comparisons lack error bars, number of runs, dataset sizes per subject, or statistical tests, so it is impossible to assess whether outperformance on cattle or the ordering across species is reliable rather than noise.

    Authors: We agree that the lack of error bars, details on the number of runs, per-subject dataset sizes, and statistical tests makes it difficult to evaluate the reliability of the reported figures and comparisons. The current results reflect the primary experimental configuration without repeated independent trials. In the revised manuscript we will rerun the key experiments across multiple random seeds, report means with standard deviations (including error bars in tables and figures), provide per-subject image counts, and include appropriate statistical comparisons where the data permit. revision: yes

  2. Referee: [Experiments / Discussion] Experiments / Discussion: the manuscript states that image quality (resolution, blur, pose) decreases from dogs to cattle to primates and that performance tracks this ordering, yet provides no ablation (resolution-matched subsets, from-scratch training on augmented animal data, or quality-controlled baselines) to isolate the contribution of FaceNet/ViT pre-training from dataset difficulty; this directly affects the transfer-learning claim.

    Authors: We acknowledge that the absence of targeted ablations limits the strength of claims isolating pre-training benefits from dataset difficulty. The manuscript already notes the quality ordering and supports the transfer approach via direct SOTA comparisons (where ad-hoc models are trained on the animal data). Full from-scratch training remains impractical given the small per-subject sample sizes emphasized in the introduction. In revision we will expand the discussion to explicitly characterize the observed quality-performance relationship as correlational, clarify the evidential limits of the transfer claim, and add a quality-controlled baseline experiment if computationally feasible within the existing data constraints. revision: partial

Circularity Check

0 steps flagged

No circularity: purely empirical transfer learning evaluation on held-out animal data

full rationale

The paper reports direct experimental results from applying pre-trained FaceNet and ViT models to three animal face datasets, measuring verification accuracy and Rank-1 identification rates, then comparing those numbers to external SOTA ad-hoc networks. No equations, parameter fits, or derivations are present; performance figures are measured outputs on the test splits rather than quantities defined or predicted from the inputs themselves. Comparisons to prior work cite independent external methods without load-bearing self-citations that would reduce the central claim to an unverified premise. The evaluation is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No free parameters, axioms, or invented entities are introduced; the work relies on standard pre-trained networks and empirical evaluation.

pith-pipeline@v0.9.1-grok · 5868 in / 1025 out tokens · 26168 ms · 2026-06-27T16:54:22.886255+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of Beyond Humans: Multispecies Animal Face Recognition Using Transfer Learning." pith.science (2026). https://pith.science/paper/ODP7YXFS

@misc{pith2026260609353,
  author       = {Pith},
  title        = {Pith review of: Beyond Humans: Multispecies Animal Face Recognition Using Transfer Learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ODP7YXFS}},
  note         = {Machine review of arXiv:2606.09353}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Individual animal recognition can be useful in the search for lost or stolen pets, the tracking of individuals of endangered species, and the recognition of animals in crowded farms. Present recognition techniques mostly use physical devices, e.g., microchips, often impractical and difficult to apply. These could be replaced by remote recognition via the animal's face; if accurate enough, it provides several advantages: it is non-invasive, can work at a distance, and is difficult to counterfeit, as, for instance, in the case of substituting sick animals for healthy ones in the food industry. The few existing datasets with sufficient per-subject images annotated with a single animal identity are not large enough to train current deep learning architectures. We rather investigate the possibility of transfer learning, exploiting pre-trained network models as backbones. Our experiments compared FaceNet, which is specifically trained on large databases of human faces, with the Vision Transformer (ViT) pre-trained on ImageNet, i.e., on object categories. We used three face datasets of very different animals: dogs, primates (lemurs, golden monkeys, and chimpanzees), and cattle. We report the results and, for each dataset, compare them with the state of the art (SOTA) ad hoc-trained deep networks. The capture conditions differ among the three datasets. Image quality (resolution, motion blur, diverse poses, etc.) decreases from dogs to cattle to primates. The best performance was achieved with dogs, where ViT reached a mean verification accuracy of 96.85% and a Rank-1 Identification Rate of 84.34%. The results for endangered primates are still encouraging, but performance varies across animal classes and tasks (verification or identification), and does not always outperform SOTA. For cattle, the ViT results outperform SOTA, while FaceNet is still competitive.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

45 extracted references · 6 canonical work pages · 1 internal anchor

  1. [1]

    S.Z. Li, A.K. Jain,Handbook of face recognition, vol. 2 (Springer, 2011)

    2011

  2. [2]

    Wanyonyi, T

    D. Wanyonyi, T. Celik, Open-source face recognition frameworks: A review of the landscape. IEEE Access10, 50601–50623 (2022)

    2022

  3. [3]

    H. Du, H. Shi, D. Zeng, X.P. Zhang, T. Mei, The elements of end-to-end deep face recognition: A survey of recent advances. ACM Computing Surveys (CSUR) 54(10s), 1–42 (2022)

    2022

  4. [4]

    Mishra, M

    N.K. Mishra, M. Dutta, S.K. Singh, Multiscale parallel deep cnn (mpdcnn) archi- tecture for the real low-resolution face recognition for surveillance. Image and Vision Computing115, 104290 (2021)

    2021

  5. [5]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

    F. Schroff, D. Kalenichenko, J. Philbin,FaceNet: A unified embedding for face recognition and clustering, in2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)(IEEE, 2015). https://doi.org/10.1109/cvpr.2015. 7298682. URL https://doi.org/10.1109%2Fcvpr.2015.7298682

    work page doi:10.1109/cvpr.2015 2015

  6. [6]

    O’toole, K.A

    A.J. O’toole, K.A. Deffenbacher, D. Valentin, H. Abdi, Structural aspects of face recognition and the other-race effect. Memory & Cognition22, 208–224 (1994)

    1994

  7. [7]

    B. Yoon, H. So, J. Rhee, A methodology for utilizing vector space to improve the performance of a dog face identification model. Applied Sciences11(5) (2021). https://doi.org/10.3390/app11052074. URL https://www.mdpi.com/2076-3417/ 11/5/2074

    work page doi:10.3390/app11052074 2021

  8. [8]

    D. Deb, S. Wiper, S. Gong, Y. Shi, C. Tymoszek, A. Fletcher, A. Jain,Face Recognition: Primates in the Wild, in2018 IEEE 9th International Conference on Biometrics Theory, Applications and Systems (BTAS)(2018), pp. 1–10. https: //doi.org/10.1109/BTAS.2018.8698538

    work page doi:10.1109/btas.2018.8698538 2018

  9. [9]

    J. Deng, W. Dong, R. Socher, L.J. Li, K. Li, L. Fei-Fei,ImageNet: A Large-Scale Hierarchical Image Database, inCVPR09(2009) 35

    2009

  10. [10]

    De Marsico, A.K

    M. De Marsico, A.K. Jain, M. Miranda, A. Orlando,Adapting to the Wild: From Human Face to Animal Face Recognition, inInternational Conference on Computer Analysis of Images and Patterns(Springer, 2025), pp. 53–64

    2025

  11. [11]

    Wilber, W.J

    M.J. Wilber, W.J. Scheirer, P. Leitner, B. Heflin, J. Zott, D. Reinke, D.K. Delaney, T.E. Boult,Animal recognition in the mojave desert: Vision tools for field biologists, in2013 IEEE Workshop on Applications of Computer Vision (WACV)(IEEE, 2013), pp. 206–213

    2013

  12. [12]

    Begon, C.R

    M. Begon, C.R. Townsend,Ecology: from individuals to ecosystems(John Wiley & Sons, 2020)

    2020

  13. [13]

    Lennox, P

    R.J. Lennox, P. Afonso, K. Birnie-Gauvin, L.S. Dahlmo, C.I. Nilsen, R. Arling- haus, S.J. Cooke, A.T. Souza, I. Jari´ c, M. Prchalov´ a, et al., Electronic tagging and tracking aquatic animals to understand a world increasingly shaped by a changing climate and extreme weather events. Canadian Journal of Fisheries and Aquatic Sciences81(3), 326–339 (2023)

    2023

  14. [14]

    Delisle, E.A

    Z.J. Delisle, E.A. Flaherty, M.R. Nobbe, C.M. Wzientek, R.K. Swihart, Next- generation camera trapping: systematic review of historic trends suggests keys to expanded research applications in ecology and conservation. Frontiers in Ecology and Evolution9, 617996 (2021)

    2021

  15. [15]

    Mizroch, J.A

    S.A. Mizroch, J.A. Beard, M. Lynde, Computer assisted photo-identification of humpback whales. Report of the International Whaling Commission12, 63–70 (1990)

    1990

  16. [16]

    L. Hiby, P. Lovell, Computer aided matching of natural markings: a prototype system for grey seals. Report of the International Whaling Commission12, 57–61 (1990)

    1990

  17. [17]

    Whitehead, Computer assisted individual identification of sperm whale flukes

    H. Whitehead, Computer assisted individual identification of sperm whale flukes. Reports of the International Whaling Commission12, 71–77 (1990)

    1990

  18. [18]

    Ardovini, L

    A. Ardovini, L. Cinque, E. Sangineto, Identifying elephant photos by multi-curve matching. Pattern Recognition41(6), 1867–1877 (2008)

    2008

  19. [19]

    Schneider, G.W

    S. Schneider, G.W. Taylor, S. Linquist, S.C. Kremer, Past, present and future approaches using computer vision for animal re-identification from camera trap data. Methods in Ecology and Evolution10(4), 461–470 (2019)

    2019

  20. [20]

    Shinoda, K

    R. Shinoda, K. Shiohara,Petface: A large-scale dataset and benchmark for animal identification, inEuropean Conference on Computer Vision(Springer, 2024), pp. 19–36

    2024

  21. [21]

    Y. Guan, Y. Lei, Y. Zhu, T. Li, Y. Xiang, P. Dong, R. Jiang, J. Luo, A. Huang, Y. Fan, et al., Face recognition of a lorisidae species based on computer vision. 36 Global Ecology and Conservation45, e02511 (2023)

    2023

  22. [22]

    Crouse, R.L

    D. Crouse, R.L. Jacobs, Z. Richardson, S. Klum, A. Jain, A.L. Baden, S.R. Tecot, Lemurfaceid: A face recognition system to facilitate individual identification of lemurs. Bmc Zoology2(1), 1–14 (2017)

    2017

  23. [23]

    Mougeot, D

    G. Mougeot, D. Li, S. Jia,A Deep Learning Approach for Dog Face Verification and Recognition, inPRICAI 2019: Trends in Artificial Intelligence, ed. by A.C. Nayak, A. Sharma (Springer International Publishing, Cham, 2019), pp. 418–430

    2019

  24. [24]

    J. Gao, T. Burghardt, W. Andrew, A.W. Dowsey, N.W. Campbell, Towards self-supervision for video identification of individual holstein-friesian cattle: The cows2021 dataset. arXiv preprint arXiv:2105.01938 (2021)

    work page arXiv 2021

  25. [25]

    Bakhshayeshi, E

    I. Bakhshayeshi, E. Erfani, F.R. Taghikhah, S. Elbourn, A. Beheshti, M. Asadnia, An intelligence cattle reidentification system over transport by siamese neural networks and yolo. IEEE Internet of Things Journal11(2), 2351–2363 (2023)

    2023

  26. [26]

    L. Hu, X. Li, G. Li, Z. Wang, Angusrecnet: Multi-module cooperation for facial anti-occlusion recognition in single-stage angus cattle. Computers and Electronics in Agriculture236, 110456 (2025)

    2025

  27. [27]

    L. Hu, X. Li, Z. Zhang, G. Li, A two-stage framework for cattle face re- identification with feature-contour and fine-texture enhancement. Journal of King Saud University Computer and Information Sciences37(8), 234 (2025)

    2025

  28. [28]

    Ahmed, J

    S.U. Ahmed, J. Frnda, M. Waqas, M.H. Khan, Dataset of cattle biometrics through muzzle images. Data in Brief p. 110125 (2024)

    2024

  29. [29]

    P. Liu, J. Zhao, Part-attention-based pseudo-label refinement reciprocal compact loss for unsupervised cattle face recognition. Electronics14(12), 2343 (2025)

    2025

  30. [30]

    Golestaneh, M

    P. Golestaneh, M. Taheri, J. Lederer,How many samples are needed to train a deep neural network?, inInternational Conference on Learning Representations, vol. 2025 (2025), pp. 63833–63858

    2025

  31. [31]

    Triantafillou, R

    E. Triantafillou, R. Zemel, R. Urtasun, Few-shot learning through an information retrieval lens. Advances in neural information processing systems30(2017)

    2017

  32. [32]

    A. Loos, A. Ernst,Detection and Identification of Chimpanzee Faces in the Wild, in2012 IEEE International Symposium on Multimedia(2012), pp. 116–119. https://doi.org/10.1109/ISM.2012.30

    work page doi:10.1109/ism.2012.30 2012

  33. [33]

    Freytag, E

    A. Freytag, E. Rodner, M. Simon, A. Loos, H.S. K¨ uhl, J. Denzler,Chimpanzee faces in the wild: Log-euclidean CNNs for predicting identities and attributes of primates, inPattern Recognition: 38th German Conference, GCPR 2016, Han- nover, Germany, September 12-15, 2016, Proceedings 38(Springer, 2016), pp. 51–63 37

    2016

  34. [34]

    W. Liu, Y. Wen, Z. Yu, M. Li, B. Raj, L. Song,Sphereface: Deep hyper- sphere embedding for face recognition, inProceedings of the IEEE conference on computer vision and pattern recognition(2017), pp. 212–220

    2017

  35. [35]

    S. Xie, R. Girshick, P. Doll´ ar, Z. Tu, K. He,Aggregated residual transformations for deep neural networks, inProceedings of the IEEE conference on computer vision and pattern recognition(2017), pp. 1492–1500

    2017

  36. [36]

    Zhang, X

    X. Zhang, X. Zhou, M. Lin, J. Sun,Shufflenet: An extremely efficient convolu- tional neural network for mobile devices, inProceedings of the IEEE conference on computer vision and pattern recognition(2018), pp. 6848–6856

    2018

  37. [37]

    F. Wang, J. Cheng, W. Liu, H. Liu, Additive margin softmax for face verification. IEEE Signal Processing Letters25(7), 926–930 (2018)

    2018

  38. [38]

    Szegedy, W

    C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Van- houcke, A. Rabinovich,Going deeper with convolutions, inProceedings of the IEEE conference on computer vision and pattern recognition(2015), pp. 1–9

    2015

  39. [39]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A.N. Gomez, L. Kaiser, I. Polosukhin, Attention is all you need. Advances in neural information processing systems30(2017)

    2017

  40. [40]

    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, N. Houlsby. An image is worth 16x16 words: Transformers for image recognition at scale (2020). https://doi.org/10.48550/ARXIV.2010.11929. URL https://arxiv.org/abs/2010. 11929

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2010.11929 2020

  41. [41]

    Nguyen, D.T

    X.B. Nguyen, D.T. Bui, C.N. Duong, T.D. Bui, K. Luu,Clusformer: A trans- former based clustering approach to unsupervised large-scale face and visual landmark recognition, inProceedings of the IEEE/CVF conference on computer vision and pattern recognition(2021), pp. 10847–10856

    2021

  42. [42]

    R. Jin, S. Zhao, Z. Hao, Y. Xu, T. Xu, E. Chen,Avt: Au-assisted visual trans- former for facial expression recognition, in2022 IEEE International Conference on Image Processing (ICIP)(IEEE, 2022), pp. 2661–2665

    2022

  43. [43]

    Devlin, M.W

    J. Devlin, M.W. Chang, K. Lee, K. Toutanova,Bert: Pre-training of deep bidi- rectional transformers for language understanding, inProceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, volume 1 (long and short papers) (2019), pp. 4171–4186

    2019

  44. [44]

    Jain, A.A

    A.K. Jain, A.A. Ross, K. Nandakumar, T. Swearingen,Introduction to Biometrics - Second edition(Springer Cham, 2025) 38

    2025

  45. [45]

    Dessimoz, C

    D. Dessimoz, C. Champod, inHandbook of biometrics(Springer, 2008), pp. 425– 459 39

    2008