Pith. sign in

REVIEW 3 major objections 5 minor 2 cited by

Self-supervised Learning on Camera Trap Footage Yields a Strong Universal Face Embedder

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A self-supervised Vision Transformer trained on unlabeled camera-trap video identifies individual chimpanzees better than supervised face-recognition baselines on a wild benchmark.

desk verdict Solid empirical case that self-supervised DINOv2 on mined camera-trap faces beats supervised baselines for wild chimpanzee re-ID, with a few fixable reporting gaps and one load-bearing data-split claim that needs auditing. read the letter →

arxiv 2507.10552 v1 pith:LPI4BFSA submitted 2025-07-14 cs.CV cs.AIcs.LG

Vladimir Iashin , Horace Lee , Dan Schofield , Andrew Zisserman This is my paper · ORCID
classification cs.CVcs.AIcs.LG
keywords self-supervisedlearningDINOv2chimpanzeefacerecognitionopen-setre-identificationcameratrapswildlifemonitoringVisionTransformer
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

Camera traps generate far more wildlife footage than experts can label, and individual identification is the main bottleneck in turning that footage into population estimates. This paper claims that identity labels are unnecessary: training a Vision Transformer with the DINOv2 self-supervised objective on roughly 964,000 automatically mined, unlabeled chimpanzee face crops yields an open-set face embedder that outperforms fully supervised baselines on the Bossou-9 wild re-identification benchmark (82.2% versus 56.7% class-averaged $k$-NN accuracy) and sets a new high on the PetFaceC* verification benchmark. The result matters because it points to a scalable, non-invasive way to monitor wild populations: point a camera trap at a forest, mine the faces, and learn to tell individuals apart without any expert annotation. The paper also shows that adding more unlabeled footage from new sites consistently improves identification accuracy.

What carries the argument

Two components carry the argument. The first is the DINOv2 self-distillation objective: a student network is trained to match the softmax outputs of a momentum-updated teacher across global and local image crops, using centering and sharpening to prevent representation collapse and patch-level alignment plus a uniformity regularizer to shape the embedding space for retrieval. The second is the Data Engine, a modular pipeline that detects faces with a lightweight detector trained on about 2,000 manually boxed frames, links detections into tracks with a generic tracker, and applies confidence filtering to keep high-quality crops while preserving track diversity; identity labels never enter this pipeline. At test time the model is used as a frozen encoder, and identification is just cosine-similarity retrieval with a $k$-NN vote.

What would settle it

Audit the split by matching every crop in the Bossou-14 training pool against the 35 tracks of the nine evaluation individuals (via track identifiers or embedding similarity); if any evaluation-individual crop is found, remove all such crops, retrain the 87M model with the same recipe, and check whether Bossou-9 re-ID still beats the 56.7% supervised baseline.

Watch

Extended reading notes

Core claim

The paper's central claim is that a universal, open-set chimpanzee face embedder can be learned with no identity labels at all. The authors train 22M- and 87M-parameter Vision Transformers from scratch with the DINOv2 objective on face crops mined from three wild video corpora, then evaluate open-set identification by $k$-nearest-neighbour search in the embedding space. On Bossou-9, the 22M model reaches 78.1% class-averaged re-ID accuracy and the 87M model 82.2%, against 56.7% for the strongest supervised baseline; on the cleaned PetFaceC* captive-chimpanzee benchmark, the 87M model achieves 76.3% verification AUC, again ahead of the supervised baselines, while its re-ID accuracy of 45.9% remains below the best supervised 49.4%. The authors interpret this as the first demonstration that a purely self-supervised, label-free training pipeline can produce a universal face embedder for a non-human species, with the retrieval-friendly geometry of DINOv2 features carrying the performance.

Load-bearing premise

The load-bearing premise is that the tracks of the nine Bossou individuals held out for evaluation were fully removed before training, so no face crop from an evaluation individual appears in the Bossou-14 training set.

Editorial extensions

If this is right

  • Adding unlabeled footage from a new site to the training mixture raises Bossou-9 re-ID from 74.6% to 78.1%, so more raw video directly translates into better identification skill without labels.
  • Scaling from a 22M- to an 87M-parameter backbone adds 4.1 points on Bossou-9 re-ID and improves verification, indicating that larger models and longer schedules are a clear route to stronger embeddings.
  • The DINOv2 self-distillation objective outperforms a contrastive baseline trained on the same data by 16.0 points on Bossou-9 re-ID, so the choice of self-supervised objective is a genuine source of the gain, not just the data volume.
  • Since the detector needs only about 2,000 boxed frames and no identity expertise, the same pipeline can be pointed at a new camera-trap archive or a new species at very low annotation cost.

Reading between the lines

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

  • Editorial extension: because nothing in the pipeline uses chimpanzee-specific identity labels, the same recipe should transfer to other species with distinctive faces or markings; a direct test is training on unlabeled footage of a new species and evaluating on a small hand-labelled set.
  • Editorial extension: the paper does not use the metadata that comes free with camera-trap footage (time, site, track continuity); treating crops from the same short track as positive pairs in an auxiliary contrastive loss is a testable way to tighten the embedding further.
  • Editorial implication: if the embedding is as robust as reported, one could cluster a large unlabeled archive first and label only cluster representatives, turning landscape-scale individual ID into a much smaller few-shot problem; the paper does not run that experiment.
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 proposes a fully self-supervised pipeline for learning chimpanzee face embeddings from unlabeled camera-trap footage. A data engine detects and tracks faces in PanAf-20K, Loma Mountains, and Bossou-14 videos, producing about 964k face crops. The authors train ViT-S and ViT-B models with the DINOv2 objective without identity labels and evaluate open-set re-identification and verification on Bossou-9 and PetFaceC*. They report that their DINOv2 models beat supervised baselines in three of four settings, establishing a new state of the art on Bossou-9 re-identification and PetFaceC* verification. The paper includes ablations comparing SimCLR to DINOv2, training-mixture contributions, and model-scaling behavior.

Significance. If the results hold, this is a valuable demonstration that self-supervised representation learning can replace identity-supervised training for animal re-identification, with clear practical benefits for conservation monitoring. The work also contributes a large mined face corpus and a reproducible data-engine pipeline, and the ablations are informative: the SimCLR-versus-DINOv2 comparison and the L/P/B training-mixture analysis directly support the central claims. However, the paper's headline numbers contain internal inconsistencies between the text and Table 3, and the Bossou-14/Bossou-9 split integrity is asserted rather than demonstrated. These issues are load-bearing for the flagship 'universal embedder' claim, so the paper needs revision before the results can be fully accepted.

major comments (3)
  1. [§5.2 vs. Table 3] The main-results text and Table 3 report different numbers for the same configurations. For the 87M ViT-B (L+P+B), §5.2 states 81.6% Bossou-9 re-ID and 74.9% verification, whereas Table 3 gives 82.2% and 74.2%. For PetFaceC*, §5.2 states 43.9% re-ID and 76.4% verification, whereas Table 3 gives 45.9% and 76.3%. Because the paper's headline claims are these numerical results, the discrepancies must be resolved and the table and text reconciled before the results can be assessed.
  2. [§4, Bossou-14/Bossou-9 split] The claim that the Bossou-14 training set is clean of evaluation identities rests on the sentence 'we first remove every track that belongs to the nine individuals reserved for evaluation.' This removal is only as reliable as the underlying identity labels and track associations; a multi-individual frame, a tracking error, or an imperfect label could place a Bossou-9 individual in the training set. Since the B-inclusive models are the flagship universal embedder and the +3.5pp gain from adding B is cited as evidence that self-supervision exploits extra video, this split needs an audit (for example, checking that no evaluation-identity face crops survive in Bossou-14, or releasing the track-to-identity mapping). Without that audit, the specific SOTA claim on Bossou-9 is not fully supported, although the L+P-only result (74.6%) partially mitigates the broad thesis.
  3. [§4, error bars and k selection] The re-ID and verification results are averaged over 10 random splits, but Table 3 reports no standard deviations or ranges, so the reader cannot judge whether differences between models are meaningful. In addition, the protocol says 'selecting the best k on a held-out split' without specifying how that split is constructed relative to the 10 random splits; this should be described precisely to rule out any selection on the evaluation queries.
minor comments (5)
  1. [§2, references] The text cites reference [16] for EfficientNetV2 when describing MiewID-msv3, but [16] is the Swin Transformer paper; the EfficientNetV2 citation is missing or mistagged.
  2. [Figure 1] The figure labels the data engine as requiring 'low-effort annotation,' while the paper emphasizes that identity labels are eliminated. This wording is ambiguous because detector training boxes are indeed low-effort, but no identity annotation is needed; it should be rephrased to avoid implying otherwise.
  3. [§3.2] The claim that a detector trained on about 2000 manually-boxed frames captures uncommon viewpoints better than generic models such as AnyFace is plausible but not supported by any quantitative comparison; if the detector choice matters, a small experiment or at least a qualitative illustration would strengthen the statement.
  4. [§5.5] The sentence 'A larger ViT-B backbone (87M parameters) and double the optimisation steps add +4.1 pp on Bossou-9 Re-ID' is consistent with Table 3's 82.2% value but contradicts the 81.6% value stated in §5.2; this is part of the same reconciliation issue noted above.
  5. [General] The paper does not state whether code, trained checkpoints, or the mined face corpus will be released beyond the project page; an explicit availability statement would improve reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the self-supervised training and evaluation are independent; the Bossou split's reliance on prior annotations is a data-integrity assumption, not circular reasoning.

full rationale

The paper's derivation chain is empirical and self-contained. The central claim is that DINOv2 trained on automatically mined face crops, with no identity labels, outperforms supervised baselines on open-set re-identification and verification. No equation fits a parameter to the evaluation benchmarks: the DINOv2 objective is a fixed, externally established self-supervised loss; the model is trained from scratch on ~964k mined crops; and the only hyperparameters (learning rate, crop sizes, k for k-NN) are chosen by standard recipes or on a held-out split, not on Bossou-9/PetFaceC* test labels. The Bossou-14 training subset is derived from the same archive as Bossou-9, and the clean split relies on identity annotations from Schofield et al. (2019), which includes two co-authors of this paper. This is a self-citation, but it is not load-bearing in a circular sense: the identity labels are used only to remove evaluation tracks before training, and are then discarded ('All identity tags are then discarded and this unlabeled subset is used solely for pre-training'); the model never sees them during training. The evaluation protocol (k-NN with gallery/query splits, ROC-AUC on balanced pairs, results averaged over random splits) is independent of the training objective. The only non-circular caveat is that the split's correctness is an empirical data-integrity assumption, not a restatement of the result; any leakage would affect validity but would not constitute circular reasoning. No fitted parameter is renamed as a prediction, and no known result is renamed as a new one. Therefore the paper earns a score of 0.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The central claim rests on the assumption that the DINOv2 objective produces retrieval-friendly features for faces, that the auto-mined face crops are clean enough, and that the Bossou-14/Bossou-9 split prevents identity leakage. Free parameters are limited to data filtering thresholds and the k-NN neighborhood size. No new entities are introduced.

free parameters (2)
  • PanAf detection confidence filter = top 20% then random 50% subsample
    Section 3.2 chooses these thresholds by hand to balance quality and diversity; they affect the training set composition.
  • k in k-NN re-identification = selected per benchmark from {1,3,5,7,10,20,50}
    Section 4 selects the best k on a held-out split; the choice is a hyperparameter tuned for each benchmark.
assumptions (4)
  • domain assumption The DINOv2 self-distillation objective, without any identity labels, produces an embedding space where cosine similarity is a valid proxy for chimpanzee identity.
    Section 3.1 motivates DINOv2 as retrieval-friendly, but this is an empirical property; the paper's results are the evidence.
  • domain assumption Track-based removal of the nine evaluation individuals from Bossou-14 is complete and correct, ensuring no identity overlap with Bossou-9.
    Section 4 states the split procedure but does not verify that no face crops of the nine individuals remain in the training set.
  • domain assumption The mined face crops are predominantly chimpanzee faces and of sufficient quality for representation learning despite the confidence filter keeping low-quality crops.
    Section 3.2 filters by confidence and subsampling but includes blurred and occluded crops by design.
  • domain assumption The identity labels used to construct Bossou-9 and Bossou-14 are reliable (from Schofield et al. 2019).
    Section 4 uses these labels to split tracks.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Self-supervised Learning on Camera Trap Footage Yields a Strong Universal Face Embedder." pith.science (2026). https://pith.science/paper/LPI4BFSA

@misc{pith2026250710552,
  author       = {Pith},
  title        = {Pith review of: Self-supervised Learning on Camera Trap Footage Yields a Strong Universal Face Embedder},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LPI4BFSA}},
  note         = {Machine review of arXiv:2507.10552}
}
read the original abstract

Camera traps are revolutionising wildlife monitoring by capturing vast amounts of visual data; however, the manual identification of individual animals remains a significant bottleneck. This study introduces a fully self-supervised approach to learning robust chimpanzee face embeddings from unlabeled camera-trap footage. Leveraging the DINOv2 framework, we train Vision Transformers on automatically mined face crops, eliminating the need for identity labels. Our method demonstrates strong open-set re-identification performance, surpassing supervised baselines on challenging benchmarks such as Bossou, despite utilising no labelled data during training. This work underscores the potential of self-supervised learning in biodiversity monitoring and paves the way for scalable, non-invasive population studies.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. Automating Visual Recognition of Leprosy in Wild Chimpanzees

    cs.CV 2026-07 conditional novelty 6.0 of 10

    A new benchmark dataset and evaluation show that simple crop-level aggregation outperforms complex video models for automated leprosy detection in camera-trap footage of wild chimpanzees.

  2. WISE: A Multimodal Search Engine for Visual Scenes, Audio, Objects, Faces, Speech, and Metadata

    cs.IR 2026-02 conditional novelty 5.0 of 10

    WISE is an open-source multimodal search engine that retrieves images, video, audio, faces, speech, and metadata using text or media queries.

Reference graph

Works this paper leans on

39 extracted references · 37 canonical work pages · cited by 2 Pith papers

  1. [1]

    INTRODUCTION Autonomous camera-trap networks are rapidly becoming the work-horse of terrestrial ecological monitoring [1]. Unlike labour-intensive techniques such as genetic sampling, transect surveys, or long-term habituation, camera traps are inexpen- sive, unobtrusive, minimally invasive, scalable, and able to record for months in habitats otherwise ha...

    work page Pith review arXiv 2025

  2. [2]

    Freytag et al

    RELATED WORK Early chimpanzee ID studies adapted human face pipelines to small, labelled datasets: Loos & Ernst [11] fused global–local handcrafted descriptors on the captive ChimpZoo and semi- wild ChimpTaï sets, but their method required meticulous alignment and ID supervision. Freytag et al. [6] introduced Log-Euclidean CNNs that jointly predicted iden...

    work page

  3. [3]

    The approach is outlined in Figure 1

    UNIVERSAL CHIMPANZEE FACE EMBEDDER This section describes how we scale face representation learn- ing beyond the limits of conventional identity-supervised pipelines. The approach is outlined in Figure 1. §3.1 Model Design shows how a self-supervised backbone such as DI- NOv2 is adapted to mined faces to yield a universal, open-set chimpanzee embedding th...

    work page

  4. [4]

    already satisfy conditions (2) and (3) by explicitly max- imizing inter-class margins and minimizing intra-class vari- ance, though they rely on labelled data and hence lie out- side the fully self-supervised scope we target. Contrastive methods such as SimCLR [21] and MoCo v3 [22] satisfy conditions (1) and (2), but they optimise primarily for inter- ima...

    work page 2000

  5. [5]

    We mine two balanced corpora with the Data Engine in §3.2: PanAf-20K and Loma Mountains each contribute ∼314k high-quality face crops

    DATASETS AND EV ALUATION Training mixture. We mine two balanced corpora with the Data Engine in §3.2: PanAf-20K and Loma Mountains each contribute ∼314k high-quality face crops. Beyond PanAf and Loma, we also use ∼335k face crops from the annotated Bossou video archive [13], containing 23 identities. Although Bossou footage was captured with camcorders on...

    work page

  6. [6]

    EXPERIMENTS 5.1. Baselines Because no prior chimpanzee embedder is self-supervised, we compare against the strongest supervised animal face de- scriptors: (i) PetFace(ArcFace ResNet50) [15] trained on the chimpanzee subset of PetFace; (ii) MegaDescriptor (ArcFace Swin-L) [17] trained on 29 wildlife multi-species datasets; (iii) MiewID-msv3 (ArcFace Effici...

    work page

  7. [7]

    Metadata accompanying each camera-trap sequence, e.g

    SUMMARY AND EXTENSIONS Our work has demonstrated that a fully self-supervised Vision Transformer trained with the DINOv2 objective on automat- ically mined chimpanzee face tracks can act as a univer- sal, open-set face embedder that outperforms fully super- vised baselines on a challenging re-identification benchmark, Bossou, and approaches their performa...

    work page

  8. [8]

    Camera traps unveil demog- raphy, social structure, and home range of six unhabituated western chimpanzee groups in the moyen bafing national park, guinea,

    B. Debetencourt, M. M. Barry, M. Arandjelovic, C. Stephens, N. Maldonado, and C. Boesch, “Camera traps unveil demog- raphy, social structure, and home range of six unhabituated western chimpanzee groups in the moyen bafing national park, guinea,” American Journal of Primatology, 2024

    work page 2024

Show all 39 references
  1. [9]

    Automatic identi- fication of individual primates with deep learning techniques,

    S. Guo, P. Xu, Q. Miao, G. Shao, C. A. Chapman, X. Chen, G. He, D. Fang, H. Zhang, Y . Sun, et al., “Automatic identi- fication of individual primates with deep learning techniques,” iScience, 2020

    2020

  2. [10]

    CosFace: Large margin cosine loss for deep face recognition,

    H. Wang, Y . Wang, Z. Zhou, X. Ji, D. Gong, J. Zhou, Z. Li, and W. Liu, “CosFace: Large margin cosine loss for deep face recognition,” in CVPR, 2018

    2018

  3. [11]

    ArcFace: Additive angular margin loss for deep face recognition,

    J. Deng, J. Guo, N. Xue, and S. Zafeiriou, “ArcFace: Additive angular margin loss for deep face recognition,” inCVPR, 2019

    2019

  4. [12]

    MS-Celeb-1M: A dataset and benchmark for large-scale face recognition,

    Y . Guo, L. Zhang, Y . Hu, X. He, and J. Gao, “MS-Celeb-1M: A dataset and benchmark for large-scale face recognition,” in ECCV, 2016

    2016

  5. [13]

    Chimpanzee faces in the wild: Log-euclidean CNNs for predicting identities and attributes of primates,

    A. Freytag, E. Rodner, M. Simon, A. Loos, H. S. Kühl, and J. Denzler, “Chimpanzee faces in the wild: Log-euclidean CNNs for predicting identities and attributes of primates,” in GCPR, 2016

    2016

  6. [14]

    Face recognition ven- dor test 2002 performance metrics,

    P. Grother, R. Micheals, and P. Phillips, “Face recognition ven- dor test 2002 performance metrics,” in International confer- ence on audio-and video-based biometric person authentica- tion, 2003

    2002

  7. [15]

    Toward open set recognition,

    Walter J Scheirer, Anderson de Rezende Rocha, Archana Sap- kota, and Terrance E Boult, “Toward open set recognition,” IEEE TPAMI, 2012

    2012

  8. [16]

    Emerging properties in self- supervised vision transformers,

    M. Caron, H. Touvron, I. Misra, H. Jégou, J. Mairal, P. Bo- janowski, and A. Joulin, “Emerging properties in self- supervised vision transformers,” in ICCV, 2021

    2021

  9. [17]

    DINOv2: Learning robust visual features without su- pervision,

    M. Oquab, T. Darcet, T. Moutakanni, H. V o, M. Szafraniec, V . Khalidov, P. Fernandez, D. Haziza, F. Massa, A. El-Nouby, et al., “DINOv2: Learning robust visual features without su- pervision,” arXiv, 2023

    2023

  10. [18]

    An automated chimpanzee identifica- tion system using face detection and recognition,

    A. Loos and A. Ernst, “An automated chimpanzee identifica- tion system using face detection and recognition,” EURASIP Journal on Image and Video Processing, 2013

    2013

  11. [19]

    Face recognition: Primates in the wild,

    D. Deb, S. Wiper, S. Gong, Y . Shi, C. Tymoszek, A. Fletcher, and A. K. Jain, “Face recognition: Primates in the wild,” in BTAS, 2018

    2018

  12. [20]

    Chimpanzee face recog- nition from videos in the wild using deep learning,

    D. Schofield, A. Nagrani, A. Zisserman, M. Hayashi, T. Mat- suzawa, D. Biro, and S. Carvalho, “Chimpanzee face recog- nition from videos in the wild using deep learning,” Science Advances, 2019

    2019

  13. [21]

    Deep residual learning for image recognition,

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

    2016

  14. [22]

    Petface: A large-scale dataset and benchmark for animal identification,

    R. Shinoda and K. Shiohara, “Petface: A large-scale dataset and benchmark for animal identification,” in ECCV, 2024

    2024

  15. [23]

    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,” in ICCV, 2021

    2021

  16. [24]

    Wildlife- datasets: An open-source toolkit for animal re-identification,

    V . ˇCermák, L. Picek, L. Adam, and K. Papafitsoros, “Wildlife- datasets: An open-source toolkit for animal re-identification,” in WACV, 2024

    2024

  17. [25]

    Otarashvili, “Miewid,” 2023

    L. Otarashvili, “Miewid,” 2023

    2023

  18. [26]

    Gorillavision – open-set re- identification of wild gorillas,

    L. Laskowski, R. Sawahn, M. Schall, D. Wasmuht, M. Bermejo, and G. de Melo, “Gorillavision – open-set re- identification of wild gorillas,” in CamTrap WS ’23 Workshop, 2023

    2023

  19. [27]

    The FAISS library,

    M. Douze, A. Guzhva, C. Deng, J. Johnson, G. Szilvasy, P.-E. Mazaré, M. Lomeli, L. Hosseini, and H. Jégou, “The FAISS library,” arXiv, 2024

    2024

  20. [28]

    A simple framework for contrastive learning of visual representations,

    T. Chen, S. Kornblith, M. Norouzi, and G. Hinton, “A simple framework for contrastive learning of visual representations,” in ICML, 2020

    2020

  21. [29]

    An empirical study of training self-supervised vision transformers,

    X. Chen, S. Xie, and K. He, “An empirical study of training self-supervised vision transformers,” in ICCV, 2021

    2021

  22. [30]

    Masked autoencoders are scalable vision learners,

    K. He, X. Chen, S. Xie, Y . Li, P. Dollár, and R. Girshick, “Masked autoencoders are scalable vision learners,” in CVPR, 2022

    2022

  23. [31]

    iBOT: Image BERT pre-training with online tok- enizer,

    J. Zhou, C. Wei, H. Wang, W. Shen, C. Xie, A. Yuille, and T. Kong, “iBOT: Image BERT pre-training with online tok- enizer,” ICLR, 2022

    2022

  24. [32]

    AnyFace: A data-centric approach for input-agnostic face detection,

    A. Kuzdeuov, D. Koishigarina, and H. A. Varol, “AnyFace: A data-centric approach for input-agnostic face detection,” in BigComp, 2023

    2023

  25. [33]

    YOLOX: Exceeding YOLO series in 2021,

    Z. Ge, S. Liu, F. Wang, Z. Li, and J. Sun, “YOLOX: Exceeding YOLO series in 2021,” arXiv, 2021

    2021

  26. [34]

    ByteTrack: Multi-object tracking by associating every detection box,

    Y . Zhang, P. Sun, Y . Jiang, D. Yu, F. Weng, Z. Yuan, P. Luo, W. Liu, and X. Wang, “ByteTrack: Multi-object tracking by associating every detection box,” in ECCV, 2022

    2022

  27. [35]

    PanAf20k: A large video dataset for wild ape detection and behaviour recognition,

    O. Brookes, M. Mirmehdi, C. Stephens, S. Angedakin, K. Corogenes, D. Dowd, P. Dieguez, T. C. Hicks, S. Jones, K. Lee, et al., “PanAf20k: A large video dataset for wild ape detection and behaviour recognition,” IJCV, 2024

    2024

  28. [36]

    Reaffirming the loma mountains national park in sierra leone as a critical site for the conservation of west african chimpanzee (pan troglodytes verus),

    G. Molina-Vacas, R. Muñoz-Mas, B. Amarasekaran, and R. M. Garriga, “Reaffirming the loma mountains national park in sierra leone as a critical site for the conservation of west african chimpanzee (pan troglodytes verus),”American Journal of Pri- matology, 2023

    2023

  29. [37]

    C3-DINO: Joint contrastive and non- contrastive self-supervised learning for speaker verification,

    C. Zhang and D. Yu, “C3-DINO: Joint contrastive and non- contrastive self-supervised learning for speaker verification,” IEEE Journal of Selected Topics in Signal Processing, 2022

    2022

  30. [38]

    Self-supervised pre-training with contrastive and masked autoencoder methods for dealing with small datasets in deep learning for medical imaging,

    D. Wolf, T. Payer, C. S. Lisson, C. G. Lisson, M. Beer, M. Götz, and T. Ropinski, “Self-supervised pre-training with contrastive and masked autoencoder methods for dealing with small datasets in deep learning for medical imaging,”Scientific Reports, 2023

    2023

  31. [39]

    Vic-MAE: Self- supervised representation learning from images and video with contrastive masked autoencoders,

    J. Hernandez, R. Villegas, and V . Ordonez, “Vic-MAE: Self- supervised representation learning from images and video with contrastive masked autoencoders,” in ECCV, 2024

    2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.