Pith. sign in

REVIEW 5 major objections 5 minor 36 references

CleanPatrick: A Benchmark for Image Data Cleaning

T0 review · 5 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read CleanPatrick is the first dedicated image-domain benchmark for data cleaning, built from 496,377 crowd-sourced judgments of off-topic images, near duplicates, and label errors in a real dermatology dataset.

desk verdict A genuinely useful first large-scale image data-cleaning benchmark, but the DINO-generated near-duplicate ground truth biases the headline comparison and needs scrutiny before the benchmark is treated as neutral. read the letter →

arxiv 2505.11034 v2 pith:RTFKQRIP submitted 2025-05-16 cs.CV cs.AIcs.LG

classification cs.CVcs.AIcs.LG
keywords datacleaningbenchmarknear-duplicatedetectionlabelerroroff-topicdermatologyimagesitemresponsetheorycrowdsourcedannotations
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

CleanPatrick is an attempt to give image-data cleaning the kind of standardized testbed that structured-data cleaning already has. The paper repurposes the 16,577-image Fitzpatrick17k dermatology collection, gathers 496,377 binary judgments from 933 medical crowd workers about three quality issues — off-topic images, near duplicates, and label errors — and converts the noisy votes into expert-verified ground truth with an item-response-theory aggregation model. The result is a benchmark that reports 4% off-topic samples, 21% near duplicates, and 22% label errors in a widely used real dataset, and that recasts each detection problem as a ranking task scored with AUROC, average precision, and top-k precision under review budgets. It matters because cleaning methods are currently compared on synthetic corruption that may not resemble real noise, whereas CleanPatrick measures them on contamination that actually occurs in an uncurated medical collection. Its main findings are that self-supervised representations nearly solve duplicate detection, classical anomaly detectors are competitive for off-topic detection under tight review budgets, and detecting clearly wrong labels in fine-grained medical classification remains unsolved.

What carries the argument

The benchmark runs on two mechanisms. The fast-duplicates procedure avoids the $O(N^2)$ cost of judging every image pair: each image is embedded with a DINO self-supervised encoder, each round sends only every image's nearest neighbor to annotators, positively confirmed pairs are merged into clusters, and the process iterates; under the assumption that an image's duplicates are always its closest neighbors, Lemma 1 guarantees that all duplicate groups are discovered in at most $\lfloor \log_2 K \rfloor + 1$ rounds using at most $N$ pair judgments, where $K$ is the size of the largest group. The aggregation model is a modified GLAD item-response-theory model that treats annotator expertise and sample difficulty as latent variables and estimates, from the sparse binary votes, a posterior probability per image that the issue is present; the sign and magnitude of the difficulty parameter encode class and confidence, and expert review calibrates one threshold per issue type. These two pieces turn noisy crowd votes into the benchmark ground truth, and the ranking formulation (AUROC, AP, P@k, R@k) is what makes any two cleaning methods comparable on the same real data.

What would settle it

Take a random subsample of several thousand Fitzpatrick17k images, annotate all image pairs within small candidate clusters using the same crowd protocol but without the nearest-neighbor pre-filter, and check whether any near-duplicate component exists in which an image's DINO nearest neighbor is not a duplicate of it; if such components appear at a non-negligible rate, the 21% duplicate figure and the duplicate rankings are biased toward DINO-like similarity. Re-running the same check with a different self-supervised embedding would show how large that bias is.

Watch

Extended reading notes

Core claim

The central claim is that a large, expert-verified contamination testbed for image cleaning is feasible, and that its construction exposes a sharp split in current methods. On CleanPatrick, the holistic method SelfClean reaches an AUROC of 0.92 and an average precision of 0.88 for near duplicates, with perfect precision among the top 1,000 ranked pairs, while perceptual hashing and SSIM stay near chance; classical anomaly detectors (Isolation Forest, HBOS, ECOD) reach AUROC 0.76–0.77 for off-topic detection and outperform richer methods on precision among the top 100 candidates; and for label errors every tested method performs at or barely above the 22% base rate. The authors conclude that self-supervised representations excel at duplicate detection, off-topic detection is serviceable when review budgets are small, and label-error detection in fine-grained medical classification remains an open problem that will need context-aware signals.

Load-bearing premise

The near-duplicate ground truth rests on the assumption that for every image, all of its true duplicates are closer to it in the paper's self-supervised embedding space than any non-duplicate, which is what lets the annotators judge only nearest-neighbor pairs; if that assumption fails, whole duplicate groups are never discovered and every duplicate-detection result is scored against an incomplete target set.

Editorial extensions

If this is right

  • Because 21% of the dataset consists of near duplicates, models trained on un-cleaned Fitzpatrick17k carry inflated class counts and a real risk of train/test leakage; the benchmark's duplicate ground truth makes both effects measurable and removable.
  • Self-supervised ranking alone places every true duplicate within the top 1,000 pairs (P@1000 = 1.00), so duplicate removal before training can be automated with very high precision.
  • Under a review budget of 100 images, SelfClean finds the most truly off-topic images (P@100 = 0.52) despite worse global ranking than the classical detectors, showing that the best method depends on whether the audit is budget-constrained.
  • Because label-error detection is at or near base rate, automated cleaning pipelines for fine-grained medical data cannot yet be trusted to find mislabels, and the paper leaves this as an open challenge.
  • Standardized ranking metrics on one real, expert-verified dataset allow cleaning methods to be compared directly against each other, replacing incompatible synthetic corruption protocols.

Reading between the lines

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

  • The near-duplicate ground truth is itself selected by a DINO embedding, so methods built on the same self-supervised representation family are being scored on positive examples tailored to that family's notion of similarity; re-deriving the duplicate labels with a second, independent embedding and re-running the benchmark would reveal how much of the 0.92 AUROC comes from sharing the selection rep
  • The instruction to mark only 'clearly wrong' labels deliberately excludes ambiguous and uncertain diagnoses, so the 22% figure is a conservative floor; a variant benchmark that scores label confidence rather than clear error would likely surface substantially more mislabels and might discriminate better among cleaning algorithms.
  • Nothing in the pipeline — task decomposition, fast-duplicates selection, GLAD aggregation, expert thresholding — is specific to dermatology, so the same procedure applied to other image collections would test whether the difficulty ordering observed here (duplicates easy, off-topic medium, labels hard) is a general property of real image contamination.
  • Because the final thresholds were calibrated by only three dermatologists, the published positive sets encode one narrow clinical judgment; re-calibrating thresholds with a larger, more diverse panel of experts is the most direct robustness check on every prevalence number in the paper.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The paper introduces CleanPatrick, which it claims is the first large-scale benchmark for image data cleaning. The benchmark is built on the Fitzpatrick17k dermatology dataset and provides binary ground-truth labels for three data-quality issues: off-topic samples, near-duplicate images, and label errors. The labels are derived from 496,377 crowd annotations by 933 medical crowd workers, aggregated with a GLAD-based item-response-theory model and calibrated with expert review of a stratified subsample. The paper formalizes each issue as a ranking task and evaluates several existing cleaning methods, reporting that SelfClean excels at near-duplicate detection (AP 0.88), classical anomaly detectors are competitive for off-topic detection, and label-error detection remains close to chance. The dataset, annotations, and evaluation code are released publicly.

Significance. If the benchmark's ground truth is trustworthy and method-neutral, CleanPatrick would be a valuable community resource: it is built on real, naturally occurring contamination rather than synthetic noise; the annotation campaign is unusually large; the ranking-task formulation matches practical audit workflows; and the release of data and code supports reproducibility and future comparison. These strengths are real and should be credited. However, the benchmark's central claim to provide a neutral testbed is currently weakened by the way the near-duplicate ground truth was constructed and by several unresolved consistency issues in the reported statistics. The near-duplicate candidate set is generated with a DINO embedding, and the method that achieves the headline near-duplicate results, SelfClean, uses closely related self-supervised representations; without validation that the candidate-generation step does not bias the evaluation, the main comparative conclusion is not yet established.

major comments (5)
  1. [§3.2, §3.5, Appendix G, Table 2] The near-duplicate ground truth is not exhaustive: candidate pairs are generated by DINO nearest-neighbor search under Assumption 2 (Fast cleaning) in Appendix G, and only those candidates are annotated. The paper provides no validation of this assumption, and Appendix A explicitly acknowledges that it may lead to incomplete discovery of near-duplicate groups. Because §3.5 states that unannotated pairs are not treated as negatives, the near-duplicate evaluation is restricted to the DINO-generated candidate set. SelfClean uses self-supervised embeddings closely related to DINO, so its reported AP of 0.88 versus pHash/SSIM near chance in Table 2 and Figure 3 may reflect the candidate-generation process rather than a general superiority in duplicate detection. To support the benchmark's neutrality, the authors should validate Assumption 2, for example by comparing candidate recall against a second independent embedding or perceptual hash and by manually inspecting a random sample of non-candidate pairs, and they should report how the evaluation changes when additional duplicate pairs found by that validation are added to the ground truth.
  2. [Abstract vs. §4.1] The abstract supplied with the submission states a 32% label-error rate, while the full-text abstract and Section 4.1 report 22% (3,666 images). The label-error prevalence is a headline quantity of the benchmark, and the inconsistency between the two versions of the abstract is a factual error that must be resolved in the final version.
  3. [§3.4] The expert-calibrated thresholds t_OT=0.76, t_ND=0.70, and t_LE=0.77 are derived from only 400 expert-annotated samples per issue type, using a rule described only as 'the bin where the distribution of positive labels starts to increase.' The final ground-truth labels, the reported prevalences, and all evaluation numbers depend directly on these thresholds, yet no sensitivity analysis is provided. The paper should specify the threshold-selection rule precisely, report how the prevalence and the benchmark results change under reasonable alternative thresholds, and ideally provide uncertainty estimates for the threshold-dependent quantities.
  4. [§2 vs. §3.2] Section 2 states that the authors 'do not use any tools or methods for performing the data selection but only rely on exhaustive annotation,' but Section 3.2 and Appendix G describe a DINO-embedding-based selection of near-duplicate candidate pairs. This is an internal contradiction about a central design choice. The near-duplicate annotation is not exhaustive, and the claim of unbiased, tool-free annotation should be corrected and qualified accordingly.
  5. [Abstract, §1, §3.4] The paper describes the released ground truth as 'expert-verified,' but expert review was performed on only 400 samples per issue type for quality control and threshold calibration; the final labels for the remaining samples come from crowd annotations aggregated with GLAD and thresholded. This overstates the level of expert verification of the released labels and should be reworded to describe the actual procedure: expert review of a stratified subsample used to calibrate and validate crowd-derived labels.
minor comments (5)
  1. [Table 2 vs. §4.1] Table 2 reports p+ = 3.7% for off-topic and 21.4% for near duplicates, while Section 4.1 reports 4% and 21%; please clarify whether p+ is computed on the evaluation subset or the full dataset and make the denominators consistent.
  2. [Abstract] The submission header abstract lists FINE and BHN among the benchmarked methods, but the full text, Appendix D, and Table 2 do not describe or report results for these methods; the abstract should be aligned with the actual experimental content.
  3. [§3.5] The positive criterion for the off-topic task says the image is off-topic 'as identified by the medical crowd workers,' but the final labels are obtained from GLAD aggregation plus expert-calibrated thresholds; the criterion should reference the actual label-generation process.
  4. [§3.2 and Appendix G] The main text says the fast-duplicates procedure requires at most 2N pairwise judgments, while Lemma 1 in Appendix G states at most |D| sample pairs; the relationship between these bounds (for example, whether each pair is annotated once or twice) should be reconciled.
  5. [§3.2] The statement that each sample received an average of 10 votes is only correct if the 496,377 annotations are divided across three issue-specific samples per image; the text should state this explicitly to avoid confusion.

Circularity Check

1 steps flagged · score 5.0 of 10

Near-duplicate ground-truth candidates are generated by DINO nearest-neighbor search, so the benchmark's headline SelfClean near-duplicate advantage is partly an artifact of candidate generation.

  1. fitted input called prediction [Section 3.2, Near duplicates; evaluation setup in Section 3.5, Task 2; assumptions in Appendix G.]
    "each image xi is embedded with an encoder that was pre-trained on ImageNet with self-supervision (i.e., DINO [20]). Its nearest neighbor n(xi) is then retrieved, and only the at most N unordered pairs {xi, n(xi)} are sent to annotation by crowd workers."

    The fast-duplicates procedure defines the candidate pool: DINO retrieves n(xi) for every image, and only those pairs are annotated. Section 3.5 then evaluates methods only on 'the annotated samples' and explicitly does not treat unannotated pairs as negatives, so the positive set available for ranking is exactly the DINO-nearest-neighbor-discoverable set. SelfClean (Appendix D.4) scores near-duplicates by nearest-neighbor distances in a learned embedding space, i.e., the same kind of similarity that generated that candidate pool.

full rationale

CleanPatrick's off-topic and label-error tasks are based on independent crowd and expert judgments and are evaluated on image-level or image-plus-label inputs, so no circularity appears there. The label-error threshold is calibrated from expert annotations, but this is a standard aggregation step, not a fitted input that is then 'predicted'. The self-citations (SelfClean [4], dermatology-benchmark work [12]) are used as related work or consistency checks, not as the justification for the reported numbers; the results are computed on this paper's own human-verified benchmark. The one load-bearing circular step is in near-duplicate evaluation: the positive candidate set was produced by DINO nearest-neighbor search under Assumption 2, and the method highlighted as excelling (SelfClean) is itself an embedding-based nearest-neighbor scorer. Because unannotated pairs are excluded from the metric computation, the near-duplicate AP/AUROC comparison measures, to a substantial degree, how well each method ranks the DINO-generated candidate pool. This is a partial circularity affecting one of three benchmark tasks; it does not make the whole benchmark vacuous, but it should be reported and ideally mitigated by validating Assumption 2 or adding exhaustive validation subsets.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central benchmark rests on a crowd-annotation pipeline and three modeling choices: the GLAD aggregation model, the fast-cleaning assumption for duplicate discovery, and expert threshold calibration. The most fragile is the fast-cleaning assumption, because it defines which duplicate pairs enter the ground truth. No invented entities are introduced.

free parameters (4)
  • Expert-calibrated threshold t_OT = 0.76
    Threshold for off-topic positive class chosen from the 400-sample expert review subset by looking for the probability bin where positive labels start to increase (Section 3.4). This threshold is fit to expert annotations, not derived.
  • Expert-calibrated threshold t_ND = 0.70
    Threshold for near-duplicate positive class chosen from the same bin-based expert review procedure (Section 3.4).
  • Expert-calibrated threshold t_LE = 0.77
    Threshold for label-error positive class chosen from the bin-based expert review procedure (Section 3.4).
  • GLAD prior scale sigma_b = 10^3
    Vague prior for item difficulty in the aggregation model (Section 3.3). Chosen by hand, not data-derived, but relatively uninformative.
assumptions (5)
  • domain assumption Fast-cleaning assumption (Assumption 2, Appendix G): for every image i, every near-duplicate j of i is closer to i in embedding space than any non-duplicate k.
    Central to discovering duplicate pairs with at most 2N annotations; if false, duplicates are missed and benchmark ground truth is incomplete. Paper acknowledges this in Appendix A.
  • domain assumption GLAD model assumption: probability of a correct annotation depends multiplicatively on annotator ability and item difficulty, Pr(y=1|c,b)=sigma(c b).
    Section 3.3. This models crowd votes; if wrong, aggregated labels are biased. It is an unproved modeling assumption.
  • domain assumption Non-adversarial majority: most annotators are not adversarial, so the sign of b_i encodes the latent class.
    Section 3.3, predictive aggregation via sign of b_i. If a sizable fraction are adversarial, labels are distorted.
  • domain assumption Expert review representativeness: 400 quantile-stratified samples per issue from three dermatologists suffice to calibrate ground-truth thresholds.
    Section 3.4. Only 3 experts and 400 samples per issue; thresholds could be idiosyncratic.
  • domain assumption DINO embedding quality: pretrained DINO on ImageNet provides a distance that reflects visual near-duplication.
    Section 3.2 and Appendix G. This underpins the fast-duplicates procedure; overlaps with the fast-cleaning assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of CleanPatrick: A Benchmark for Image Data Cleaning." pith.science (2026). https://pith.science/paper/RTFKQRIP

@misc{pith2026250511034,
  author       = {Pith},
  title        = {Pith review of: CleanPatrick: A Benchmark for Image Data Cleaning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RTFKQRIP}},
  note         = {Machine review of arXiv:2505.11034}
}
read the original abstract

Robust machine learning depends on clean data, yet current image data cleaning benchmarks rely on synthetic noise or narrow human studies, limiting comparison and real-world relevance. We introduce CleanPatrick, the first large-scale benchmark for data cleaning in the image domain, built upon the publicly available Fitzpatrick17k dermatology dataset. We collect 496,377 binary annotations from 933 medical crowd workers, identify off-topic samples (4%), near-duplicates (21%), and label errors (32%), and employ an aggregation model inspired by item-response theory followed by expert review to derive high-quality ground truth. CleanPatrick formalizes issue detection as a ranking task and employs standard ranking metrics that mirror real audit workflows. We benchmark classical anomaly detectors, perceptual hashing, SSIM, Confident Learning, NoiseRank, FINE, BHN, and SelfClean. On CleanPatrick, self-supervised representations excel at near-duplicate detection, classical methods achieve competitive off-topic detection under constrained review budgets, and detecting implausible labels under conservative human judgment remains challenging for fine-grained medical classification. By releasing both the dataset and the evaluation framework, CleanPatrick enables a systematic comparison of image-cleaning strategies.

Figures

Figures reproduced from arXiv: 2505.11034 by the authors.

Figure 1
Figure 1. Process of acquiring and curating the CleanPatrick benchmark. We started by collecting [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Examples of data quality issues identified in the Fitzpatrick17k dataset. [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Performance of different data cleaning approaches (represented in colors) for the three [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Left shows a screenshot of the labeling interface shown to the medical crowd workers. [PITH_FULL_IMAGE:figures/full_fig_p013_4.png]
Figure 5
Figure 5. Figure 5: Histograms showing the number of annotations from medical crowd workers per image [PITH_FULL_IMAGE:figures/full_fig_p014_5.png]
Figure 6
Figure 6. Figure 6: Distribution of per-sample annotation uncertainty, measured as [PITH_FULL_IMAGE:figures/full_fig_p015_6.png]
Figure 7
Figure 7. Figure 7: Number of duplicates per batch of annotation. For each batch, we select the closest pairs [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]
Figure 8
Figure 8. Figure 8: Inter-annotator agreement as Krippendorff’s alpha among all expert annotators (left) and [PITH_FULL_IMAGE:figures/full_fig_p017_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 29 canonical work pages

  1. [1]

    Dinov2: Learning robust visual features without supervision.arXiv preprint arXiv:2304.07193, 2023

    Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, et al. Dinov2: Learning robust visual features without supervision.arXiv preprint arXiv:2304.07193, 2023

  2. [2]

    Revisiting Automatic Data Curation for Vision Foundation Models in Digital Pathology

    Boqi Chen, Cédric Vincent-Cuaz, Lydia A Schoenpflug, Manuel Madeira, Lisa Fournier, Vaishnavi Subramanian, Sonali Andani, Samuel Ruiperez-Campillo, Julia E V ogt, Raphaëlle Luisier, et al. Re- visiting Automatic Data Curation for Vision Foundation Models in Digital Pathology.arXiv preprint arXiv:2503.18709, 2025

  3. [3]

    Northcutt, Anish Athalye, and Jonas Mueller

    Curtis G. Northcutt, Anish Athalye, and Jonas Mueller. Pervasive Label Errors in Test Sets Destabilize Machine Learning Benchmarks. InAdvances in Neural Information Processing Systems, 2021

  4. [4]

    Navarini, and Marc Pouly

    Fabian Gröger, Simone Lionetti, Philippe Gottfrois, Alvaro Gonzalez-Jimenez, Ludovic Amruthalingam, Labelling Consortium, Matthew Groh, Alexander A. Navarini, and Marc Pouly. Intrinsic Self-Supervision for Data Quality Audits.Advances in Neural Information Processing Systems, 2024

  5. [5]

    Confident learning: Estimating uncertainty in dataset labels

    Curtis Northcutt, Lu Jiang, and Isaac Chuang. Confident learning: Estimating uncertainty in dataset labels. Journal of Artificial Intelligence Research, 2021

  6. [6]

    Noiserank: Unsupervised label noise reduction with dependence models

    Karishma Sharma, Pinar Donmez, Enming Luo, Yan Liu, and I Zeki Yalniz. Noiserank: Unsupervised label noise reduction with dependence models. InEuropean Conference on Computer Vision, 2020

  7. [7]

    Rein: A comprehensive benchmark framework for data cleaning methods in ml pipelines.arXiv preprint arXiv:2302.04702, 2023

    Mohamed Abdelaal, Christian Hammacher, and Harald Schoening. Rein: A comprehensive benchmark framework for data cleaning methods in ml pipelines.arXiv preprint arXiv:2302.04702, 2023

  8. [8]

    Cleanml: A study for evaluating the impact of data cleaning on ml classification tasks

    Peng Li, Xi Rao, Jennifer Blase, Yue Zhang, Xu Chu, and Ce Zhang. Cleanml: A study for evaluating the impact of data cleaning on ml classification tasks. InInternational Conference on Data Engineering, 2021

Show all 36 references
  1. [9]

    Dc-bench: Dataset condensation benchmark

    Justin Cui, Ruochen Wang, Si Si, and Cho-Jui Hsieh. Dc-bench: Dataset condensation benchmark. Advances in Neural Information Processing Systems, 2022

  2. [10]

    Dataperf: Benchmarks for data-centric ai development.URL https://arxiv

    Mark Mazumder, Colby Banbury, Xiaozhe Yao, Bojan Karlaš, William Gaviria Rojas, Sudnya Diamos, Greg Diamos, Lynn He, Douwe Kiela, David Jurado, et al. Dataperf: Benchmarks for data-centric ai development.URL https://arxiv. org/abs/2207.10062, 2022

  3. [11]

    Evaluating deep neural networks trained on clinical images in dermatology with the fitzpatrick 17k dataset

    Matthew Groh, Caleb Harris, Luis Soenksen, Felix Lau, Rachel Han, Aerin Kim, Arash Koochek, and Omar Badri. Evaluating deep neural networks trained on clinical images in dermatology with the fitzpatrick 17k dataset. InConference on Computer Vision and Pattern Recognition, 2021

  4. [12]

    Towards reliable dermatology evaluation benchmarks

    Fabian Gröger, Simone Lionetti, Philippe Gottfrois, Alvaro Gonzalez-Jimenez, Matthew Groh, Roxana Daneshjou, Alexander A Navarini, Marc Pouly, Labelling Consortium, et al. Towards reliable dermatology evaluation benchmarks. InMachine Learning for Health, 2023

  5. [13]

    Investigating the quality of dermamnist and fitzpatrick17k dermatological image datasets.Scientific Data, 2025

    Kumar Abhishek, Aditi Jain, and Ghassan Hamarneh. Investigating the quality of dermamnist and fitzpatrick17k dermatological image datasets.Scientific Data, 2025

  6. [14]

    Aqua: A benchmarking tool for label quality assessment.Advances in Neural Information Processing Systems, 2023

    Mononito Goswami, Vedant Sanil, Arjun Choudhry, Arvind Srinivasan, Chalisa Udompanyawit, and Artur Dubrawski. Aqua: A benchmarking tool for label quality assessment.Advances in Neural Information Processing Systems, 2023

  7. [15]

    Human uncertainty makes classification more robust

    Joshua C Peterson, Ruairidh M Battleday, Thomas L Griffiths, and Olga Russakovsky. Human uncertainty makes classification more robust. InInternational Conference on Computer Vision, 2019. 10

  8. [16]

    Benchmarking unsupervised near-duplicate image detection.Expert Systems with Applications, 2019

    Lia Morra and Fabrizio Lamberti. Benchmarking unsupervised near-duplicate image detection.Expert Systems with Applications, 2019

  9. [17]

    Natural Adversarial Examples.Conference on Computer Vision and Pattern Recognition, 2021

    Dan Hendrycks, Kevin Zhao, Steven Basart, Jacob Steinhardt, and Dawn Song. Natural Adversarial Examples.Conference on Computer Vision and Pattern Recognition, 2021

  10. [18]

    Roxana Daneshjou, Mert Yuksekgonul, Zhuo Ran Cai, Roberto Novoa, and James Y . Zou. SkinCon: A skin disease dataset densely annotated by domain experts for fine-grained debugging and analysis. In Advances in Neural Information Processing Systems, 2022

  11. [19]

    A General-Purpose Multimodal Foundation Model for Dermatology.arXiv preprint arXiv:2410.15038, 2024

    Siyuan Yan, Zhen Yu, Clare Primiero, Cristina Vico-Alonso, Zhonghua Wang, Litao Yang, Philipp Tschandl, Ming Hu, Gin Tan, Vincent Tang, et al. A General-Purpose Multimodal Foundation Model for Dermatology.arXiv preprint arXiv:2410.15038, 2024

  12. [20]

    Emerging properties in self-supervised vision transformers

    Mathilde Caron, Hugo Touvron, Ishan Misra, Hervé Jégou, Julien Mairal, Piotr Bojanowski, and Armand Joulin. Emerging properties in self-supervised vision transformers. InInternational Conference on Computer Vision, 2021

  13. [21]

    Whose V ote Should Count More: Optimal Integration of Labels from Labelers of Unknown Expertise

    Jacob Whitehill, Ting-fan Wu, Jacob Bergsma, Javier Movellan, and Paul Ruvolo. Whose V ote Should Count More: Optimal Integration of Labels from Labelers of Unknown Expertise. InAdvances in Neural Information Processing Systems, 2009

  14. [22]

    Chen, Martin Jankowiak, Fritz Obermeyer, Neeraj Pradhan, Theofanis Karaletsos, Rohit Singh, Paul A

    Eli Bingham, Jonathan P. Chen, Martin Jankowiak, Fritz Obermeyer, Neeraj Pradhan, Theofanis Karaletsos, Rohit Singh, Paul A. Szerlip, Paul Horsfall, and Noah D. Goodman. Pyro: Deep Universal Probabilistic Programming.J. Mach. Learn. Res., 2019

  15. [23]

    Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization.arXiv preprint arXiv:1412.6980, 2014

  16. [24]

    DSM-5 field trials in the United States and Canada, Part II: test-retest reliability of selected categorical diagnoses.American journal of psychiatry, 2013

    Darrel A Regier, William E Narrow, Diana E Clarke, Helena C Kraemer, S Janet Kuramoto, Emily A Kuhl, and David J Kupfer. DSM-5 field trials in the United States and Canada, Part II: test-retest reliability of selected categorical diagnoses.American journal of psychiatry, 2013

  17. [25]

    Handling inter-annotator agreement for automated skin lesion segmentation.arXiv preprint arXiv:1906.02415, 2019

    Vinicius Ribeiro, Sandra Avila, and Eduardo Valle. Handling inter-annotator agreement for automated skin lesion segmentation.arXiv preprint arXiv:1906.02415, 2019

  18. [26]

    Comparison of visual diagnostic accuracy of dermatologists practicing in Germany in patients with light skin and skin of color.Scientific reports, 2024

    Frederik Krefting, Maurice Moelleken, Stefanie Hölsken, Jan-Malte Placke, Robin Tamara Eisenburger, Lea Jessica Albrecht, Alpaslan Tasdogan, Dirk Schadendorf, Selma Ugurel, Joachim Dissemond, et al. Comparison of visual diagnostic accuracy of dermatologists practicing in Germa...

  19. [27]

    Watermarking images in self-supervised latent spaces

    Pierre Fernandez, Alexandre Sablayrolles, Teddy Furon, Hervé Jégou, and Matthijs Douze. Watermarking images in self-supervised latent spaces. InIEEE International Conference on Acoustics, Speech and Signal Processing, 2022

  20. [28]

    Isolation Forest

    Fei Tony Liu, Kai Ming Ting, and Zhi-Hua Zhou. Isolation Forest. InIEEE International Conference on Data Mining, 2008

  21. [29]

    Histogram-based outlier score (hbos): A fast unsupervised anomaly detection algorithm.KI-2012: poster and demo track, 2012

    Markus Goldstein and Andreas Dengel. Histogram-based outlier score (hbos): A fast unsupervised anomaly detection algorithm.KI-2012: poster and demo track, 2012

  22. [30]

    ECOD: Unsupervised Outlier Detection Using Empirical Cumulative Distribution Functions.IEEE Transactions on Knowledge and Data Engineering, 2022

    Zheng Li, Yue Zhao, Xiyang Hu, Nicola Botta, Cezar Ionescu, et al. ECOD: Unsupervised Outlier Detection Using Empirical Cumulative Distribution Functions.IEEE Transactions on Knowledge and Data Engineering, 2022

  23. [31]

    D. Marr, E. Hildreth, and Sydney Brenner. Theory of edge detection.Proceedings of the Royal Society of London. Series B. Biological Sciences, 1997

  24. [32]

    Bovik, H.R

    Zhou Wang, A.C. Bovik, H.R. Sheikh, and E.P. Simoncelli. Image quality assessment: from error visibility to structural similarity.IEEE Transactions on Image Processing, 2004

  25. [33]

    A short introduction to boosting.Journal-Japanese Society For Artificial Intelligence, 1999

    Yoav Freund, Robert Schapire, and Naoki Abe. A short introduction to boosting.Journal-Japanese Society For Artificial Intelligence, 1999

  26. [34]

    O jistém problému minimálním.Práce Mor

    Otakar Bor˚ uvka. O jistém problému minimálním.Práce Mor. Prırodved. Spol. v Brne (Acta Societ. Scienc. Natur. Moravicae), 1926

  27. [35]

    Efficient algorithms for mining outliers from large data sets

    Sridhar Ramaswamy, Rajeev Rastogi, and Kyuseok Shim. Efficient algorithms for mining outliers from large data sets. InACM SIGMOD International Conference on Management of Data, 2000

  28. [36]

    Mean Structural Similarity Index

    Fabrizio Angiulli and Clara Pizzuti. Fast outlier detection in high dimensional spaces. InEuropean conference on principles of data mining and knowledge discovery, 2002. 11 A Limitations Our benchmark acknowledges three limitations: (1) By presenting annotators with only each ...

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.