Pith. sign in

REVIEW 3 major objections 4 minor 54 references

Automated screening of camera-trap footage can flag visible leprosy in wild chimpanzees; simple pooling of per-crop predictions matches or beats complex video models because the disease is a static skin condition.

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 · deepseek-v4-flash

2026-08-01 18:04 UTC pith:VVLZN2MB

load-bearing objection First ML-ready leprosy dataset and a sensible benchmark, but the headline ordering (simple pooling > video models; trim/top-α helps) rests on 17 positive test tracks and test-set-tuned α, so treat the rankings as provisional. the 3 major comments →

arxiv 2607.17395 v1 pith:VVLZN2MB submitted 2026-07-19 cs.CV

Automating Visual Recognition of Leprosy in Wild Chimpanzees

Katie I. Murray , Anna C. Bowland , Marina Ramon , Elena Bersacola , Aissa Regalla , Manmohan D. Sharma , Markus Mueller , Majid Mirmehdi
show 4 more authors
Dave Hodgson Kimberley J. Hockings Tilo Burghardt Otto Brookes
This is my paper
classification cs.CV
keywords leprosy detectionwildlife disease monitoringcamera trapschimpanzeesdeep learningvideo classificationtemporal aggregationPanLep300
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.

Leprosy in wild western chimpanzees produces visible skin signs, but camera-trap networks generate far more footage than experts can review. The paper introduces PanLep300, a dataset of 125,670 annotated bounding-box crops from 953 tracks in 303 videos spanning five years in Guinea-Bissau, and uses it to test whether deep learning can screen for leprosy automatically. Its central finding is that simple aggregation of per-crop predictions—mean probability or majority vote—consistently matches or outperforms both learned temporal aggregators and end-to-end video models, consistent with leprosy being a static cutaneous condition rather than a motion-dependent one. The paper also identifies a practical failure mode: tracklets containing partially visible individuals at track start and end dilute confident interior predictions, and this can be corrected by trimming boundary frames or averaging only the top-scoring fractions. If correct, automated screening becomes feasible with relatively simple image classifiers plus probability pooling, and the same approach could extend to other visually identifiable wildlife diseases.

Core claim

On the paper's own terms, the discovery is that explicit localisation plus simple probability pooling solves the problem: crop-level classification of individual chimpanzees far outperforms full-frame classification, and averaging or majority-voting per-crop probabilities across a track beats trained temporal models and end-to-end video architectures at both tracklet and full-track evaluation. At full-track level, mean probability and majority vote reach 0.941 balanced accuracy with perfect precision on the 256-test-track set (15 of 17 positive tracks caught, no false alarms). The paper further shows that a systematic 'boundary dilution' effect—partially visible individuals at track entry/ex

What carries the argument

The load-bearing mechanism is crop-level classification with probability pooling: each frame is first reduced to a tight bounding-box crop around an individual, a frozen image classifier scores each crop, and track-level decisions are made by pooling those per-crop probabilities (mean, max, or majority vote) rather than by learning temporal dynamics. The paper's ecologically motivated split—withholding whole named individuals and entire camera locations from training—is what makes the comparison trustworthy. A second mechanism is tracklet construction: padding incomplete clips versus trimming remainder frames, plus top-alpha percent mean aggregation, which together counteract the boundary-di

Load-bearing premise

The central comparisons rest on only 17 positive test tracks from a single national park and on visual-only labels, so if those tracks are unrepresentative or if visible signs do not track infection, the reported method rankings and the gains from trimming could change.

What would settle it

Run the same pipeline on a new, larger test set with at least a few hundred positive tracks from multiple parks, seasons, and camera models, ideally with molecular confirmation. If a trained temporal aggregator or end-to-end video model then beats simple mean or majority voting by a wide margin, or if trimming and top-alpha aggregation no longer recover boundary-diluted positives, the paper's central ranking and the static-phenotype explanation would be refuted.

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

If this is right

  • Landscape-scale leprosy surveillance becomes feasible: camera-trap footage can be screened automatically, flagging tracks that need expert review rather than requiring manual review of every frame.
  • Complex 3D video models are unnecessary for static visible phenotypes; a frozen per-crop classifier plus mean or majority voting is enough, simplifying deployment and reducing compute.
  • Track construction and aggregation choices materially affect performance: trimming partial entry/exit frames and averaging only high-scoring chunks recovers positives that would otherwise be missed.
  • The same pipeline should transfer to any wildlife disease with persistent visual signs, in any species that appears in camera-trap footage, assuming enough labelled data.
  • The released dataset gives other researchers a benchmark with splits that test generalization to unseen individuals and unseen cameras.

Where Pith is reading between the lines

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

  • Because temporal models add no signal, track-level consistency could be repurposed as a confidence filter: requiring two or more independently high-scoring crops before flagging a track would likely cut false alarms further, at a small recall cost.
  • The boundary-dilution result suggests a cheap deployment fix: adjust camera-trap trigger settings or mounting to capture full-body mid-trail passages, or have the tracker discard the first and last few frames of each track as a standard preprocessing step.
  • The 'None' label means no visible signs, not no infection; a combined screening plus molecular sampling protocol could estimate the hidden early-stage reservoir that visual screening misses.
  • The background-masking result—performance drops when background is removed—raises the possibility that some signal is habitat-correlated rather than lesion-correlated; a testable next step is to compare masking variants that preserve skin regions only.

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

3 major / 4 minor

Summary. The paper introduces PanLep300, a dataset of 125,670 annotated bounding-box crops from 953 tracks across 303 camera-trap videos of wild chimpanzees, with track-level leprosy labels and splits designed to withhold whole individuals and camera installations. The authors benchmark 2D image classifiers, 2.5D temporally aggregated models, and 3D video models, and report three main findings: (1) crop-level classification with any-positive aggregation substantially outperforms full-frame classification; (2) simple hand-coded aggregation of crop probabilities (mean, majority vote) matches or beats trained temporal aggregators and end-to-end video models; and (3) padding artefacts at tracklet boundaries dilute confidence scores, and this can be mitigated by trimmed tracklet construction and top-alpha percentage aggregation. The paper releases the dataset as a public benchmark and claims the first automated deep-learning pipeline for leprosy screening in wildlife.

Significance. If the empirical findings hold, this is a valuable contribution: it provides the first public ML-ready dataset for a wildlife disease from camera-trap imagery, and the conclusion that simple crop-probability pooling suffices for a static cutaneous phenotype is practically important and consistent with evidence from human clinical AI. The ecological split design, withholding named individuals and unseen cameras, is a genuine strength for testing generalization rather than identity-memorisation. The paper is also honest in Section 5 about the small positive test set. However, the headline comparative and boundary-dilution claims currently rest on a test set of only 17 positive tracks and on aggregation hyperparameters selected on that same test set, which makes the quantitative rankings and the measured benefit of top-alpha aggregation statistically fragile. The dataset itself is a solid basis for future work; the evaluation methodology needs strengthening before the specific quantitative claims can be accepted as stated.

major comments (3)
  1. [Section 4.4, Table 6 footnote] The top-alpha percentages are chosen per architecture 'to maximise balanced accuracy' on the same 256-track test set. This is test-set selection and makes the reported top-alpha gains optimistic. With n+=17, one positive-track flip changes balanced accuracy by ~2.9 percentage points; the headline X3D-M gain (trimmed top-25% 0.964 vs mean-softmax 0.912, 0.052; vs heuristic mean 0.941, 0.023) is within the range that could arise by selecting the best of five alpha values on the evaluation data. Please select alpha on the validation set (or via nested cross-validation) and report test performance for the pre-selected alpha, or report all alpha values with bootstrap confidence intervals.
  2. [Tables 2–6] No uncertainty quantification is reported anywhere: no confidence intervals, bootstrap intervals, or repeated-seed runs. The comparative claims in Table 4 rely on differences as small as 0.004 (majority vote vs mean probability at T=32) and 0.026 (majority vote vs X3D-M at T=16), and Table 5 differences of 0.088 between heuristic and the best video model. With only 17 positive tracks, exact binomial or bootstrap intervals around balanced accuracy are essential to interpret these gaps. Please add them, at least for the track-level results in Tables 5 and 6.
  3. [Section 4.4, 'Boundary Tracklets Dilute...' paragraph] The boundary-dilution narrative is developed by inspecting per-tracklet probabilities on the 17 positive test tracks, and the trimming/top-alpha strategies are then evaluated on that same set, with alpha selected on the test set. This is a post hoc explanation and risks overfitting the narrative to the evaluation data. The paper's own Section 5 acknowledges that close comparisons are 'statistically fragile', but the abstract and conclusion state the boundary-dilution finding and its correction as established. Please either validate the proposed strategies on a held-out split or explicitly frame this part as exploratory, and temper the abstract/conclusion wording accordingly.
minor comments (4)
  1. [Table 6] The header contains a formatting artifact: 'Padded T rimmed' should be 'Padded' and 'Trimmed'.
  2. [References 30 and 42] Reference titles contain garbled characters ('2020âĂŞ2030', 'HansenâĂŹs disease'); please fix the encoding.
  3. [Section 3.3] For the individual Tasha, the text says the disease status changes across the recording period but the track-level label 'remains fixed for the duration of the track.' Please clarify how tracks are labelled when the visible status changes within a track, and how many tracks of Tasha are labelled Advanced vs None.
  4. [Tables 2 and 3] Table 2 reports frame-level evaluation of crop classification and Table 3 crop-level evaluation; the near-identical ResNet-18 rows may confuse readers. A one-sentence note distinguishing the evaluation units would help.

Circularity Check

0 steps flagged

No circular derivation: results are empirical comparisons on held-out tracks; minor post-hoc test-set selection of top-α is a methodological caveat, not circularity.

full rationale

The paper makes no first-principles derivation that reduces to its inputs. Its central comparisons (crop vs full-frame, heuristic vs trained aggregation vs 3D video models, trimmed/top-α aggregation) are evaluated on a held-out test set of 256 tracks with ecologically motivated splits that withhold named individuals and camera installations (Sec. 3.4, Tables 2, 4, 5, 6). No equation is defined in terms of the target result, and no fitted parameter is relabeled as a prediction: model weights are trained on the designated train/val splits and performance is reported on test tracks. The only caveat is that the top-α percentages in Table 6 are chosen 'to maximise balanced accuracy' on that same 256-track test set, and the boundary-dilution narrative was developed after inspecting per-tracklet test outputs; this makes the aggregation gains post-hoc selections rather than independent confirmations, and with only 17 positive test tracks the comparisons are 'statistically fragile' as the paper itself states in Sec. 5. That is an evaluation/selection limitation, not a definitional circularity: the reported improvement over mean-softmax is not forced by construction, and no self-citation is load-bearing for the method choice (domain references such as [28] and [6] provide background and are not used to derive the benchmark outcomes). Overall, the derivation chain is self-contained empirical benchmarking; score 2 reflects the minor test-set-dependent selection rather than any circular reasoning.

Axiom & Free-Parameter Ledger

1 free parameters · 3 axioms · 0 invented entities

The central claims rest on the dataset labels and the evaluation protocol rather than on new theory. The main free parameter is the test-fitted α; no new entities are introduced.

free parameters (1)
  • α (top-alpha aggregation percentage) = p:25%/t:25% for Transformer; p:33%/t:50% for Bi-LSTM; p:33%/t:25% for Mean-pool MLP; p:33%/t:25% for X3D-M; p:50%/t:25%
    Selected per architecture to maximise balanced accuracy on the 256-track test set; this is fitting to the evaluation data and inflates reported boundary-dilution fixes.
axioms (3)
  • domain assumption Expert visual assignment of 'Advanced' vs 'None' leprosy signs in camera-trap imagery is a reliable ground truth for the task.
    Labels are track-level visual attributes (Section 3.3), with molecular confirmation in only one individual cited from prior work [28]. The paper itself notes 'None' denotes absence of visible signs, not confirmed absence of infection.
  • domain assumption ImageNet and Kinetics pre-trained weights transfer sufficiently to camera-trap chimpanzee crops.
    All models are initialised with these weights (§4.1) and no domain-adaptive pre-training is done; the benchmarks assume this transfer is valid.
  • domain assumption The ecological split (disjoint individuals and camera locations) fully prevents shortcut learning of identity or camera cues.
    Split design §3.4 withholds 4 named individuals and 2 camera sites, but there may be other community-level features shared across splits; no analysis verifies absence of leakage beyond the stated rules.

pith-pipeline@v1.3.0-alltime-deepseek · 15715 in / 13466 out tokens · 125659 ms · 2026-08-01T18:04:48.740511+00:00 · methodology

0 comments
read the original abstract

Leprosy (Mycobacterium leprae) has been confirmed in wild western chimpanzees (Pan troglodytes verus) in West Africa, presenting as clear and progressive visual symptoms. Manual review of camera-trap footage at landscape scale is infeasible, motivating the need for automated screening. We present the first deep learning pipeline for wildlife leprosy detection and contribute the PanLep300 dataset of 125,670 annotated bounding-box crops across 953 tracks from 303 camera-trap videos with ecologically-motivated splits that withhold whole individuals and camera installations. We benchmark spatial (2D), temporally aggregated (2.5D), and video-based (3D) classification approaches to investigate which approach is best suited to automated leprosy detection in wild apes. We find that simple aggregation of crop-level predictions consistently matches or outperforms both learned temporal models and end-to-end video architectures -- consistent with leprosy's static cutaneous presentation. We further find that performance is suppressed when tracklets contain frames of partially visible individuals -- as commonly occurs at the start and end of a track -- and demonstrate that this can be addressed through targeted construction and aggregation strategies.

Figures

Figures reproduced from arXiv: 2607.17395 by Aissa Regalla, Anna C. Bowland, Dave Hodgson, Elena Bersacola, Katie I. Murray, Kimberley J. Hockings, Majid Mirmehdi, Manmohan D. Sharma, Marina Ramon, Markus Mueller, Otto Brookes, Tilo Burghardt.

Figure 1
Figure 1. Figure 1: Conceptual Overview. The PanLep300 Dataset comprises 125,670 annotated bounding-box crops across 71,508 frames of wild chimpanzee camera-trap footage with track-level leprosy labels. We benchmark image classifiers, temporal aggregators, and end-to-end video models surfacing actionable findings on the value of explicit localisa￾tion, the hazards of background masking, and tracklet boundary dilution. it prod… view at source ↗
Figure 2
Figure 2. Figure 2: Data Collection at Cantanhez National Park. (A) Research area and land cover within CNP [43], with the study community’s range at Caiquene￾Cadique (CC) outlined in white. (B) Camera-trap placements (light blue points) within the community’s home range, estimated via 100% MCP analysis of 1,380 chimpanzee observations (2013–2018); some points overlap due to paired or repeated placements across years. (C) Exa… view at source ↗
Figure 3
Figure 3. Figure 3: Track Distribution across Chimpanzee Individuals. Distribution and leprosy symptom status across PanLep300. (A) Number of tracks per named individ￾ual (41 identity classes), ordered by total track count and coloured by symptom status: Advanced (red; visible leprosy signs) and None (blue; no observable symptoms), plot￾ted on a log scale. Three non-individual classes – chimpanzee (115 tracks), Infant (199 tr… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative Comparison of Frame and Crop-level Classification. Test frame examples where full-frame classification and crop-level classification disagree at frame-level. Bounding boxes indicate ground-truth region localisation, coloured green (correctly localised) or red (missed). The header on each frame shows full-frame model prediction, crop-aggregated frame-level prediction, and ground-truth label. Cro… view at source ↗
Figure 5
Figure 5. Figure 5: Predicted Leprosy-likelihood Distribution for X3D-M. Data shown for Stage-A-only, trimmed T=16, top-25% aggregation on the 256-track test set. A) Box￾and-whisker plots for None (blue) and Advanced (red) leprosy in tracks; dashed line marks the 0.5 threshold. B) Log-scale histogram across 10 probability bins. The distri￾bution is strongly bimodal: 226/239 negative tracks score below 0.05 and 14/17 positive … view at source ↗
Figure 6
Figure 6. Figure 6: Grad-CAM Visualisations of Leprosy Classifier for X3D-M. Shown are four 16 frame skims Stage-A-only, trimmed T=16, at top-25% aggregation. Top Skim shows false negative with attention collapse to the foreground treculia fruit (Treculia africana) – all models fail on this track. 2nd shows true positive with near-zero activation during occlusion. 3rd shows true positive with tightly localised attention hotsp… view at source ↗

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

54 extracted references · 5 canonical work pages

  1. [1]

    PLOS Computational Biology21(6), e1012550 (2025) 3

    de Andrade, H.G.V., da Silva Rocha, E., de Carvalho Monteiro, K.H., de Morais, C.M., Moura dos Santos, D.C., Cassimiro Nascimento, D., Dourado, R.A., Lynn, T., Endo, P.T.: On the usage of artificial intelligence in leprosy care: A systematic literature review. PLOS Computational Biology21(6), e1012550 (2025) 3

    2025

  2. [2]

    Journal of Applied Ecology62(11), 2894–2913 (2025) 3, 5, 15

    Barroso, P., Palencia, P.: Pixelated pathologies: Camera trapping as a tool for monitoring wildlife health. Journal of Applied Ecology62(11), 2894–2913 (2025) 3, 5, 15

    2025

  3. [3]

    In: 2023 International Conference on Sustainable Com- puting and Data Communication Systems (ICSCDS)

    Baweja, A.K., Aditya, S., Kanchana, M.: Leprosy diagnosis using explainable artifi- cial intelligence techniques. In: 2023 International Conference on Sustainable Com- puting and Data Communication Systems (ICSCDS). pp. 551–556. IEEE (2023) 3, 15

    2023

  4. [4]

    Beery, S., Morris, D., Yang, S.: Efficient pipeline for camera trap image review (2019) 6

    2019

  5. [5]

    Bersacola, E., Hockings, K.J.: Action plan for the conservation of medium- and large-sized terrestrial mammals in Cantanhez National Park, Guinea-Bissau (2024- 2034). Tech. rep., Institute for Biodiversity and Protected Areas (IBAP), Bissau (2023) 2

    2024

  6. [6]

    Scientific Reports11(1), 4569 (2021) 4

    Bersacola, E., Hill, C.M., Hockings, K.J.: Chimpanzees balance resources and risk in an anthropogenic landscape of fear. Scientific Reports11(1), 4569 (2021) 4

    2021

  7. [7]

    Landscape Ecology37(12), 3103–3121 (2022) 4 16 K.I

    Bersacola, E., Hill, C.M., Nijman, V., Hockings, K.J.: Examining primate com- munity occurrence patterns in agroforest landscapes using arboreal and terrestrial camera traps. Landscape Ecology37(12), 3103–3121 (2022) 4 16 K.I. Murray et al

    2022

  8. [8]

    In: 2016 IEEE International Conference on Image Processing (ICIP)

    Bewley, A., Ge, Z., Ott, L., Ramos, F., Upcroft, B.: Simple online and realtime tracking. In: 2016 IEEE International Conference on Image Processing (ICIP). pp. 3464–3468. IEEE (2016).https://doi.org/10.1109/ICIP.2016.75330036

    arXiv 2016

  9. [9]

    Lancet363(9416), 1209–1219 (2004).https://doi.org/10.1016/S0140-6736(04)15952-7,https://doi.org/ 10.1016/S0140-6736(04)15952-71

    Britton, W.J., Lockwood, D.N.: Leprosy. Lancet363(9416), 1209–1219 (2004).https://doi.org/10.1016/S0140-6736(04)15952-7,https://doi.org/ 10.1016/S0140-6736(04)15952-71

    work page doi:10.1016/s0140-6736(04)15952-7 2004

  10. [10]

    In: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion

    Brookes, O., Kukushkin, M., Mirmehdi, M., Stephens, C., Dieguez, P., Hicks, T.C., Jones, S., Lee, K., McCarthy, M.S., Meier, A., et al.: The panaf-fgbg dataset: Un- derstanding the impact of backgrounds in wildlife behaviour recognition. In: Pro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recogni- tion. pp. 5433–5443 (2025) 2, 3

    2025

  11. [11]

    arXiv preprint arXiv:2301.02642 (2023) 3

    Brookes, O., Mirmehdi, M., Kühl, H., Burghardt, T.: Triple-stream deep metric learning of great ape behavioural actions. arXiv preprint arXiv:2301.02642 (2023) 3

    Pith/arXiv arXiv 2023

  12. [12]

    International Journal of Computer Vision132(8), 3086–3102 (2024) 3

    Brookes, O., Mirmehdi, M., Stephens, C., Angedakin, S., Corogenes, K., Dowd, D., Dieguez, P., Hicks, T.C., Jones, S., Lee, K., et al.: Panaf20k: a large video dataset for wild ape detection and behaviour recognition. International Journal of Computer Vision132(8), 3086–3102 (2024) 3

    2024

  13. [13]

    arXiv preprint arXiv:2603.26754 (2026) 2, 4

    Brundage, D.: Generating synthetic wildlife health data from camera trap im- agery: A pipeline for alopecia and body condition training data. arXiv preprint arXiv:2603.26754 (2026) 2, 4

    arXiv 2026

  14. [14]

    Clinical Microbiology and Infection18(6), 521–527 (2012) 1

    Calvignac-Spencer, S., Leendertz, S.A.J., Gillespie, T.R., Leendertz, F.H.: Wild great apes as sentinels and sources of infectious disease. Clinical Microbiology and Infection18(6), 521–527 (2012) 1

    2012

  15. [15]

    arXiv preprint arXiv:2511.16719 (2025) 8

    Carion, N., Gustafson, L., Hu, Y.T., Debnath, S., Hu, R., Suris, D., Ryali, C., Alwala,K.V.,Khedr,H.,Huang,A.,etal.:Sam3:Segmentanythingwithconcepts. arXiv preprint arXiv:2511.16719 (2025) 8

    Pith/arXiv arXiv 2025

  16. [16]

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

    Carreira, J., Zisserman, A.: Quo vadis, action recognition? a new model and the kinetics dataset. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). pp. 6299–6308 (2017) 7, 10

    2017

  17. [17]

    org/10.1098/rstb.2016.0167,https://doi.org/10.1098/rstb.2016.016715

    Cunningham, A.A., Daszak, P., Wood, J.L.N.: One health, emerging infectious diseases and wildlife: two decades of progress? Philosophical Transactions of the Royal Society B: Biological Sciences372(1725), 20160167 (2017).https://doi. org/10.1098/rstb.2016.0167,https://doi.org/10.1098/rstb.2016.016715

    arXiv 2017

  18. [18]

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

    Deng, J., Dong, W., Socher, R., Li, L.J., Li, K., Fei-Fei, L.: Imagenet: A large-scale hierarchical image database. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR). pp. 248–255. IEEE (2009) 7

    2009

  19. [19]

    Revista Panamericana de Salud Pública50, e40 (2026) 3

    Deps, P., Amorim, B.B.C., Repsold, T., Almonfrey, D., do Espírito Santo, R.B., Loureiro, R.M., Enechukwu, N.A., Barreiro, T.Z., Rodrigues, M.M., Fonseca, A.M.F., et al.: Independent assessment of the who skin neglected tropical diseases application for leprosy detection. Revista Panamericana de Salud Pública50, e40 (2026) 3

    2026

  20. [20]

    EcoHealth15, 148–162 (2018).https://doi.org/10.1007/s10393-017-1306-11, 4

    Dunay, E., Apakupakul, K., Leard, S., et al.: Pathogen transmission from humans to great apes is a growing threat to primate conservation. EcoHealth15, 148–162 (2018).https://doi.org/10.1007/s10393-017-1306-11, 4

    work page doi:10.1007/s10393-017-1306-11 2018

  21. [21]

    Na- ture542(7639), 115–118 (2017).https://doi.org/10.1038/nature21056,https: //doi.org/10.1038/nature210563, 15 Leprosy Recognition in Chimpanzees 17

    Esteva, A., Kuprel, B., Novoa, R.A., Ko, J., Swetter, S.M., Blau, H.M., Thrun, S.: Dermatologist-level classification of skin cancer with deep neural networks. Na- ture542(7639), 115–118 (2017).https://doi.org/10.1038/nature21056,https: //doi.org/10.1038/nature210563, 15 Leprosy Recognition in Chimpanzees 17

    doi:10.1038/nature21056 2017

  22. [22]

    In: Proceedings of the IEEE/CVF conference on computer vision and pattern recog- nition

    Feichtenhofer, C.: X3d: Expanding architectures for efficient video recognition. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recog- nition. pp. 203–213 (2020) 10

    2020

  23. [23]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Feichtenhofer, C., Fan, H., Malik, J., He, K.: Slowfast networks for video recog- nition. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 6202–6211 (2019) 10

    2019

  24. [24]

    IET Computer Vision18(8), 1193–1208 (2024) 2

    Gadot, T., Istrate, È., Kim, H., Morris, D., Beery, S., Birch, T., Ahumada, J.: To crop or not to crop: Comparing whole-image and cropped classification on a large dataset of camera trap images. IET Computer Vision18(8), 1193–1208 (2024) 2

    2024

  25. [25]

    Supplement to the best practice guidelines for health monitoring and disease control in great ape populations

    Gilardi, K.V., Düx, A., Gillespie, T.R., Kalema-Zikusoka, G., Leendertz, F., Nze- Nkogue, C., Patrono, L.V., Raphael, J., Sulistyo, F., Kambale Syaluha, E., Whit- tier, C.: An update on great ape population health, disease control and outbreak response. Supplement to the best practice guidelines for health monitoring and disease control in great ape popul...

    2025

  26. [26]

    JAMA316(22), 2402–2410 (2016).https://doi.org/10.1001/jama.2016.17216, https://doi.org/10.1001/jama.2016.172163

    Gulshan, V., Peng, L., Coram, M., Stumpe, M.C., Wu, D., Narayanaswamy, A., Venugopalan, S., Widner, K., Madams, T., Cuadros, J., Kim, R., Raman, R., Nel- son, P.C., Mega, J.L., Webster, D.R.: Development and validation of a deep learn- ing algorithm for detection of diabetic retinopathy in retinal fundus photographs. JAMA316(22), 2402–2410 (2016).https://...

    arXiv 2016

  27. [27]

    Hernandez, A., Miao, Z., Vargas, L., Beery, S., Dodhia, R., Lavista, J.: Pytorch- wildlife: A collaborative deep learning framework for conservation (2024),https: //arxiv.org/abs/2405.129306

    Pith/arXiv arXiv 2024

  28. [28]

    Nature598(7882), 652–656 (2021) 2, 4, 5, 6

    Hockings, K.J., Mubemba, B., Avanzi, C., Pleh, K., Düx, A., Bersacola, E., Bessa, J., Ramon, M., Metzger, S., Patrono, L.V., et al.: Leprosy in wild chimpanzees. Nature598(7882), 652–656 (2021) 2, 4, 5, 6

    2021

  29. [29]

    arXiv preprint arXiv:2507.10552 (2025) 3

    Iashin, V., Lee, H., Schofield, D., Zisserman, A.: Self-supervised learning on camera trap footage yields a strong universal face embedder. arXiv preprint arXiv:2507.10552 (2025) 3

    Pith/arXiv arXiv 2025

  30. [30]

    IUCN SSC Primate Specialist Group: Regional action plan for the conservation of western chimpanzees (Pan troglodytes verus) 2020âĂŞ2030. Tech. rep., IUCN, Gland, Switzerland (2020).https://doi.org/10.2305/IUCN.CH.2020.SSC-RAP. 2.en,https://doi.org/10.2305/IUCN.CH.2020.SSC-RAP.2.en1, 4, 15

    work page doi:10.2305/iucn.ch.2020.ssc-rap 2020

  31. [31]

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

    Jing, Y., Wu, J., Yang, Z., Liang, K., Zhu, X., Ma, Z.: Ethoclip: Ontology-enhanced video-language pretraining for animal behavior understanding. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 31196–31206 (2026) 3

    2026

  32. [32]

    arXiv preprint arXiv:1910.09217 (2019) 8

    Kang, B., Xie, S., Rohrbach, M., Yan, Z., Gordo, A., Feng, J., Kalantidis, Y.: Decoupling representation and classifier for long-tailed recognition. arXiv preprint arXiv:1910.09217 (2019) 8

    Pith/arXiv arXiv 1910

  33. [33]

    Trends in Ecology & Evolution28(7), 432–441 (2013) 3

    Kühl, H.S., Burghardt, T.: Animal biometrics: quantifying and detecting pheno- typic appearance. Trends in Ecology & Evolution28(7), 432–441 (2013) 3

    2013

  34. [34]

    American Journal of Primatology79(9), e22681 (2017) 1

    Kühl, H.S., Sop, T., Williamson, E.A., Mundry, R., Brugière, D., Campbell, G., Cohen, H., Danquah, E., Ginn, L., Herbinger, I., et al.: The critically endangered western chimpanzee declines by 80%. American Journal of Primatology79(9), e22681 (2017) 1

    2017

  35. [35]

    CamTrap WS23(2023) 3 18 K.I

    Laskowski, L., Sawahn, R., Schall, M., Wasmuht, D., Bermejo, M., de Melo, G.: Gorillavision–open-set reidentification of wild gorillas. CamTrap WS23(2023) 3 18 K.I. Murray et al

    2023

  36. [36]

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

    Mueller, F.B., Meier, J.F., Lueddecke, T., Vogg, R., Freixanet, R.L., Hassler, V., Bosshard, T., Karakoc, E., O’Hearn, W.J., Pereira, S.M., et al.: Privi: Towards a general-purpose video model for primate behavior in the wild. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 38804–38815 (2026) 3

    2026

  37. [37]

    The Journal of Wildlife Diseases55(4), 770–781 (2019) 2, 3

    Muneza, A.B., Ortiz-Calo, W., Packer, C., Cusack, J.J., Jones, T., Palmer, M.S., Swanson, A., Kosmala, M., Dickman, A.J., Macdonald, D.W., et al.: Quantifying the severity of giraffe skin disease via photogrammetry analysis of camera trap data. The Journal of Wildlife Diseases55(4), 770–781 (2019) 2, 3

    2019

  38. [38]

    Nature403(6772), 853–858 (2000) 4

    Myers, N., Mittermeier, R.A., Mittermeier, C.G., Da Fonseca, G.A., Kent, J.: Bio- diversity hotspots for conservation priorities. Nature403(6772), 853–858 (2000) 4

    2000

  39. [39]

    Emerging Microbes & Infections8(1), 139–149 (2019).https://doi.org/10.1080/22221751.2018

    Negrey, J.D., Reddy, R.B., Scully, E.J., Phillips-Garcia, S., Owens, L.A., Langer- graber, K.E., Goldberg, T.L.: Simultaneous outbreaks of respiratory disease in wild chimpanzees caused by distinct viruses of human origin. Emerging Microbes & Infections8(1), 139–149 (2019).https://doi.org/10.1080/22221751.2018. 15634561, 4

    arXiv 2019

  40. [40]

    Proceedings of the National Academy of Sciences115(25), E5716–E5725 (2018) 3

    Norouzzadeh, M.S., Nguyen, A., Kosmala, M., Swanson, A., Palmer, M.S., Packer, C., Clune, J.: Automatically identifying, counting, and describing wild animals in camera-trap images with deep learning. Proceedings of the National Academy of Sciences115(25), E5716–E5725 (2018) 3

    2018

  41. [41]

    Wildlife Research51(1) (2024) 2, 3

    Nurçin, F.V., Şentürk, N., Imanov, E., Thalmann, S., Fagg, K., Jiang, G.: Auto- mated tasmanian devil segmentation and devil facial tumour disease classification. Wildlife Research51(1) (2024) 2, 3

    2024

  42. [42]

    World Health Organization (2021) 1

    Organization, W.H., et al.: Towards zero leprosy: global Leprosy (HansenâĂŹs disease) Strategy 2021–2030. World Health Organization (2021) 1

    2021

  43. [43]

    Remote Sensing Applications: Society and Environ- ment26, 100746 (2022) 4

    Pereira, S.C., Lopes, C., Pedroso, J.P.: Mapping cashew orchards in cantanhez national park (guinea-bissau). Remote Sensing Applications: Society and Environ- ment26, 100746 (2022) 4

    2022

  44. [44]

    A five- group system

    Ridley, D.S., Jopling, W.H.: Classification of leprosy according to immunity. A five- group system. International Journal of Leprosy and Other Mycobacterial Diseases 34(3), 255–273 (1966) 1

    1966

  45. [45]

    Integrative Zoology21(1), 116–128 (2026) 2, 4

    Ringwaldt, E.M., Buettel,J.C., Carver,S., Brook,B.W.: Epidemiological dynamics of a visually apparent disease: Camera trapping and machine-learning applied to rumpwear in the common brushtail possum. Integrative Zoology21(1), 116–128 (2026) 2, 4

    2026

  46. [46]

    In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision

    Schall, M., Knöfel, F.L., König, N.E., Kubeler, J.J., von Klinski, M., Linnemann, J.W., Liu, X., Schlegelmilch, I.J., Woyciniuk, O., Schild, A., et al.: Gorillawatch: An automated system for in-the-wild gorilla re-identification and population mon- itoring. In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision. pp. 8364–8375...

    2026

  47. [47]

    Methods in Ecology and Evolution10(4), 461–470 (2019) 3

    Schneider, S., Taylor, G.W., Linquist, S., Kremer, S.C.: 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) 3

    2019

  48. [48]

    Science advances5(9), eaaw0736 (2019) 3

    Schofield, D., Nagrani, A., Zisserman, A., Hayashi, M., Matsuzawa, T., Biro, D., Carvalho, S.: Chimpanzee face recognition from videos in the wild using deep learning. Science advances5(9), eaaw0736 (2019) 3

    2019

  49. [49]

    Sekachev, B., Manovich, N., Zhiltsov, M., Zhavoronkov, A., Kalinin, D., Hoff, B., TOsmanov, K.D., Zankevich, A., et al.: opencv/cvat: v1. 1.0. zenodo. Melnikov, J. Kim, L. Ilouz, N. Glazov, Priya, R. Tehrani, S. Jeong, V. Skubriev, S. Yonekura, vugia truong, zliang7, lizhming and T. Truong, opencv/cvat: v11(2020) 6 Leprosy Recognition in Chimpanzees 19

    2020

  50. [50]

    Future Mi- crobiology6(10), 1151–1157 (oct 2011).https://doi.org/10.2217/fmb.11.97, https://doi.org2, 5

    Suzuki, K., Tanigawa, K., Kawashima, A., Miyamura, T., Ishii, N.: Chimpanzees used for medical research shed light on the pathoetiology of leprosy. Future Mi- crobiology6(10), 1151–1157 (oct 2011).https://doi.org/10.2217/fmb.11.97, https://doi.org2, 5

    work page doi:10.2217/fmb.11.97 2011

  51. [51]

    Conservation Biology p

    Tichon, J., Lapid, R., King, R., Freiman, E., Spiegel, O., Baruchi, E., Atar, A., Bar-David,S.,Saltz,D.:Monitoringwildlifehealthfordiseaseswithvisiblesignsby integrating camera traps with marked individuals. Conservation Biology p. e70283 (2026) 2, 3

    2026

  52. [52]

    Nature communications13(1), 792 (2022) 2, 3, 15

    Tuia,D.,Kellenberger,B.,Beery,S.,Costelloe,B.R.,Zuffi,S.,Risse,B.,Mathis,A., Mathis, M.W., Van Langevelde, F., Burghardt, T., et al.: Perspectives in machine learning for wildlife conservation. Nature communications13(1), 792 (2022) 2, 3, 15

    2022

  53. [53]

    Clinics in Dermatology25(2), 165–172 (2007).https://doi.org/10.1016/j.clindermatol.2006.05.0061

    Walker, S.L., Lockwood, D.N.: Leprosy. Clinics in Dermatology25(2), 165–172 (2007).https://doi.org/10.1016/j.clindermatol.2006.05.0061

    work page doi:10.1016/j.clindermatol.2006.05.0061 2007

  54. [54]

    arXiv preprint arXiv:2511.15622 (2025) 2, 3, 8

    Wasmuht, D.F., Brookes, O., Schall, M., Palencia, P., Beirne, C., Burghardt, T., Mirmehdi, M., Kühl, H., Arandjelovic, M., Pottie, S., et al.: The sa-fari dataset: Segment anything in footage of animals for recognition and identification. arXiv preprint arXiv:2511.15622 (2025) 2, 3, 8

    arXiv 2025