Pith. sign in

REVIEW 4 major objections 5 minor 30 references

SDQM: Synthetic Data Quality Metric for Object Detection Dataset Evaluation

T0 review · 4 major / 5 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read SDQM predicts object-detection mAP from dataset quality alone, at r=0.87.

desk verdict The integrated-metric recipe is plausible and some pieces are new, but the headline r=0.87 is inflated by selection leakage and a 10-epoch mAP ground truth; the honest number is closer to the k-fold Spearman of 0.58. read the letter →

arxiv 2510.06596 v2 pith:XD64DWWD submitted 2025-10-08 cs.CV cs.AIcs.ITcs.LGmath.IT

Ayush Zenith , Arnold Zumbrun , Neel Raut , Jing Lin This is my paper · ORCID
classification cs.CVcs.AIcs.ITcs.LGmath.IT
keywords syntheticdatadatasetqualityobjectdetectionevaluationmetricmAPdomaingapfeatureembeddingsrandomforestregression
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

SDQM is an integrated metric that aims to score how useful a synthetic image dataset will be for training an object detection model, without requiring the model to be trained to convergence. The paper combines several existing and newly designed sub-metrics—covering pixel intensities, bounding-box geometry, label statistics, spatial layout, and embedding-space overlap—and feeds them into a random forest regressor that outputs a single quality score. The central claim is that this score correlates strongly (Pearson r=0.87 on one evaluation split; 0.78±0.091 in k-fold) with the mAP50 achieved by a YOLOv11n model trained on the dataset, far above existing metrics like α-Precision, β-Recall, and MAUVE. If true, SDQM gives a cheap screening tool for synthetic data generation and selection.

What carries the argument

The central object is the SDQM score itself: a random forest regression over selected sub-metrics that maps a (real, synthetic) dataset pair to a single number. The regression is the mechanism that lifts correlation beyond any individual sub-metric, and it is trained on mAP50 values from a YOLOv11n model trained for ten epochs. The V-Information sub-metric—an extension of V-usable information to object detection—carries the most weight; it measures the change in predictive entropy of a COCO-pretrained YOLOv11n after brief fine-tuning on synthetic data. Feature embeddings for the distribution-comparison sub-metrics come from a GroundingDINO-tiny encoder, selected by cosine similarity between

What would settle it

Train object detectors to convergence (e.g., 100+ epochs) on a diverse set of real/synthetic subset pairs, compute SDQM for each pair, and measure the Pearson correlation between SDQM and converged mAP50. If the correlation drops well below 0.87—or below the correlation of a cheap baseline like β-Recall—the metric is an artifact of early-training dynamics rather than a true dataset-quality measure.

Watch

Extended reading notes

Core claim

The paper claims that dataset quality for object detection can be captured by a composite of interpretable sub-metrics, independently of a converged training run. SDQM is built from components measuring pixel-intensity match, bounding-box match, label overlap, spatial distribution difference, dataset separability, clusterability, α-Precision, β-Recall, Authenticity, MAUVE, Frontier Integral, and an extension of V-usable information called V-Information. Sub-metrics are pruned via correlation and backward feature reduction, and the survivors are combined with random forest regression trained against actual YOLOv11n mAP50 scores on synthetic/real subset pairs selected by an evolutionary algori

Load-bearing premise

The metric is validated against the mAP50 of a YOLOv11n model trained for only ten epochs on small, evolutionarily curated subset pairs, and the paper assumes that this short-training mAP orders datasets the same way as fully converged training.

Editorial extensions

If this is right

  • A team generating synthetic data can rank candidate datasets by SDQM before spending GPU-hours on training.
  • SDQM can be used as an objective in generation or selection loops, replacing costly train-and-evaluate iteration.
  • The metric's components give diagnostic feedback: a low Pixel Intensity Match or low β-Recall pinpoints why a synthetic dataset may underperform.
  • The same structure is portable to instance segmentation, localization, and classification by swapping the detector used for V-Information.
  • SDQM provides a benchmark for future research on mixed real-synthetic dataset curation.

Reading between the lines

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

  • Because the regression is trained on 10-epoch YOLOv11n mAP, the metric may be tuned to early-training behavior; its correlation with fully converged mAP is untested and could be weaker.
  • The feature-extractor selection criterion—cosine similarity between real and synthetic DIMO pairs—could be satisfied trivially by a constant encoder, so SDQM's validity depends on the encoder actually capturing semantics.
  • The reported r=0.87 comes from one train/eval split; the k-fold average is lower (0.78), so the headline correlation may overstate typical performance.
  • SDQM could be repurposed as an acquisition function for active learning or data pruning on real datasets, not just synthetic-data screening.
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

4 major / 5 minor

Summary. The paper proposes SDQM, an integrated metric for evaluating synthetic object-detection datasets without requiring converged model training. SDQM combines established metrics (MAUVE, Frontier Integral, α-Precision/β-Recall/Authenticity, V-Usable Information) with new components (Dataset Separability, Clusterability, spatial/box/pixel/label distribution matches) and fuses them via random forest regression to predict YOLOv11n mAP50. Experiments use an evolutionary algorithm to create varied subset pairs from RarePlanes, DIMO, and WASABI, yielding 364 training and 64 evaluation subset pairs. The paper reports a held-out Pearson correlation of r=0.8719 for SDQM versus mAP50, with an average k-fold Pearson correlation of 0.78±0.091, and compares SDQM against several existing scalar metrics.

Significance. If the r=0.87 result were obtained under a genuinely clean held-out protocol, SDQM would be a practically valuable cheap surrogate for training-based dataset evaluation in object detection, with clear uses in synthetic-data generation, selection, and debugging. The paper has real strengths: it releases code, uses three diverse dataset pairs, evaluates multiple feature extractors, and reports k-fold variability. However, the headline correlation is currently not trustworthy because model and feature selection are performed on the full dataset before the train/eval split, and because the ground truth is only 10-epoch mAP50, never validated against converged training. These issues are fixable within the manuscript's scope, so the central idea is defensible but needs a substantially reworked evaluation.

major comments (4)
  1. [§3.3, §4, Appendices D/E] Selection leakage invalidates the headline held-out r. Section 3.3 performs backward feature reduction on all collected data points to remove MAUVE, Authenticity, and Clusterability before the train/eval split described in Section 4. Appendix D selects the random forest regression family by comparing methods on the same full data, and Appendix E selects per-sub-metric statistical tests and quadratic/interaction terms via regressions on those same data points. The k-fold cross-validation in Section 4 reuses the pre-selected feature set and model family in every fold, so the reported r=0.8719 (Table 2) and k-fold mean 0.78±0.091 are optimistic. Please redo all feature/model/transform selection inside the training folds (nested CV) or otherwise demonstrate that the reported correlation holds on a truly untouched evaluation set.
  2. [§4, Model Training] The ground truth used throughout is mAP50 of YOLOv11n trained for only ten epochs. The paper never checks whether this 10-epoch score ranks datasets the same way as converged training. If short-training mAP is an artifact of optimization dynamics rather than dataset quality, SDQM may be predicting that artifact. Please validate the proxy by training at least a subset of the curated pairs to convergence (e.g., 50–100 epochs) and reporting rank correlation between 10-epoch and converged mAP50, or otherwise justify epoch choice.
  3. [Table 2] The comparison against existing metrics is unfair and therefore not informative as reported. SDQM is a 9-feature random forest fitted to mAP50, while the baselines are raw scalar metrics evaluated without any calibration or fitting. A more appropriate comparison would apply the same fitting protocol to each baseline — e.g., fit each scalar metric through a monotonic regression on training folds and report cross-validated Pearson/Spearman correlations with confidence intervals. As presented, Table 2 conflates the benefit of the feature set with the benefit of regression fitting.
  4. [Appendix A] The feature-extractor selection criterion is not sufficient to establish information quality. The paper selects GroundingDino-tiny because it yields high cosine similarity and low Euclidean distance between paired real and synthetic DIMO images. But a degenerate constant encoder would trivially maximize cosine similarity (all pairs identical) and minimize Euclidean distance. The criterion therefore does not measure whether the encoder preserves useful semantic information. Please add a control or a task-based probe, such as linear-probe classification accuracy or a comparison against random/constant embeddings.
minor comments (5)
  1. [Abstract/§1] The abstract and contributions state that SDQM 'sets a new standard' and 'demonstrated a strong correlation (r=0.87)' without acknowledging the k-fold variability (0.78±0.091) and the selection-leakage caveat. Please moderate the claim or qualify it with the CV estimate.
  2. [§3.1, Eq. (1)] The log cluster metric uses 'log' without a base; specify the base. Also clarify how the expression behaves when the squared term is zero or very small.
  3. [§4] The text says 'this process results in 364 training subsets and 64 evaluation subsets,' but the preceding description of the evolutionary algorithm and random additions is not fully formal. Please clarify whether the 64 evaluation subsets are disjoint from the training subsets and how the 'randomly selected' extra subsets were incorporated.
  4. [Appendix E] Appendix E states that Dataset Separability regression assigns a negative weight to validation accuracy a and a positive weight to parameter count p, then says 'this establishes a positive correlation' with mAP50 where 'higher values ... indicate greater difficulty in distinguishing.' The sign logic is unclear: higher a should mean easier separation, so the intended combination needs a more precise explanation.
  5. [Throughout] There are several typos and formatting issues: 'W ASABI' spacing, 'futhermore', 'displayes' (Appendix D), and Table 5's mixed text prompts. The code link is given, but no license or version/commit hash is provided for reproducibility.

Circularity Check

2 steps flagged · score 6.0 of 10

Feature and model selection on the full dataset before the train/eval split makes SDQM's headline r=0.87 a partially self-confirming, selection-contaminated estimate.

  1. fitted input called prediction [Sec. 3.3 (SDQM: Integrated Metric Development) and Sec. 4 (SDQM Development); Fig. 3 and Table 2]
    "After conducting an exploratory study to combine the sub-components of each sub-metric and analyzing their relationship with mAP50, we determine that random forest regression best fits the data ... We remove these three sub-metrics from the data before proceeding with the regression analysis in Sec. 4. ... After the above procedures, a regression can be performed with the sub-metric values between a subset pair as inputs and the mAP50 score of the respective YOLOv11n model as the output. These values for each dataset pair are collected and shuffled before splitting into train and evaluation se"

    The paper itself documents that the regression family (random forest) and the feature subset (MAUVE, Authenticity, Clusterability removed) were chosen by 'analyzing their relationship with mAP50' and backward feature reduction on all collected datapoints in Sec. 3.3, and only afterward were values 'collected and shuffled before splitting into train and evaluation sets' in Sec. 4. The validation points behind Fig. 3 and Table 2 therefore already influenced which SDQM features and which regression method are used. The reported r=0.8719 is a correlation between the fitted regressor's output and its own training target (mAP50), measured on a split that is not truly independent of the fitting procedure; the k-fold mean (Pearson 0.78±0.091, Spearman 0.58±0.14) reuses the same pre-selected featur

  2. fitted input called prediction [Appendix E (Sub-Metric Regression Analysis)]
    "We perform quadratic analysis to combine sub-metrics into sub-components, removing terms with p-values greater than 0.05 for statistical significance."

    Appendix E tunes each SDQM component's internal formula (which of the A-D, ED, K-S statistical tests; which quadratic terms; which interaction terms survive the p<0.05 threshold) by regressions on the collected data, with mAP50 as the target (the surrounding text links these regressions to 'predicting mAP' and to correlation with mAP50). These selected transforms are then baked into the SDQM inputs before the Sec. 4 train/eval split. The correlation reported on the validation split thus inherits a fit to those very validation points, applied at the sub-component level - the same pre-split selection leakage as in Sec. 3.3.

full rationale

SDQM is a random forest whose training target is YOLOv11n mAP50, so a correlation between SDQM output and mAP50 is, by itself, an ordinary surrogate-validation claim rather than a tautology; the sub-metric inputs are computed independently of mAP50, and the final regression is evaluated on a shuffled split, so the central claim has real content. The partial circularity lies in the order of operations, which the paper itself describes: Sec. 3.3 selects the regression family by 'analyzing their relationship with mAP50' on all collected data and removes features via backward reduction 'on the data'; Appendix E selects per-component statistical tests and quadratic/interaction terms via p-value regressions on the same data; only then, in Sec. 4, are values 'collected and shuffled before splitting into train and evaluation sets.' The validation points that produce Fig. 3 and Table 2 (r=0.8719) were therefore used, before the split, to choose which features enter SDQM, which transforms each component uses, and which regressor is reported; the k-fold mean reuses the same pre-selection, so cross-validation does not remove the contamination. The headline correlation is thus an optimistic, selection-inflated estimate of the fit's own output against its own target. Separately, the ground truth is 10-epoch YOLOv11n mAP50 (Sec. 4, Model Training) with no evidence that 10-epoch ranking matches converged training, and Table 2 compares the fitted SDQM against unfitted generic metrics; these are validity concerns, not circularity. There is no self-citation chain, so the self-citation patterns do not apply. Because the sub-metric values themselves are computed independently of the target and the final regression parameters are genuinely out-of-sample, the result is only partially self-confirming, hence 6 rather than 8-10.

Assumptions & free parameters 7 free parameters · 6 assumptions · 4 invented entities

The paper's contribution is an estimator: SDQM is a fitted function of mAP50, with feature-set and per-sub-metric test selection also made against mAP50 on the full data. The genuinely external inputs (MAUVE, α-Precision, etc.) come from prior literature, but their selection and weighting are data-fit. No new physical entities are postulated; the four 'invented' entries are methodological constructs with only in-paper evidence.

free parameters (7)
  • Random forest regression coefficients (SDQM weights) = not disclosed (no model dump or feature importances given)
    SDQM is the output of a random forest fitted to YOLOv11n mAP50 on subset pairs (Sec. 3.3–4, Fig. 3); all reported SDQM-vs-mAP correlations are the fitted model's held-out scores.
  • Feature subset choice (drop MAUVE, Authenticity, Clusterability) = removed
    Backward feature reduction on the full dataset against mAP50 (Sec. 3.3) — done before the train/eval split, so the choice is fitted on data that includes evaluation points.
  • Per-sub-metric statistical test selection = A-D (Pixel Intensity), ED (Bounding Box), K-S (Label Overlap)
    Selected by p-values from regression against mAP50 (Sec. 3.3, Appendix E); a data-dependent modeling choice.
  • V-Information fine-tuning epochs E = 10
    E=10 epochs, backbone frozen, YOLOv11n (Sec. 3.1); chosen for compute/performance balance, not justified against convergence.
  • Dataset Separability sub-component weights = unspecified ('weighted appropriately')
    Sec. 3.1 states parameter count and validation accuracy 'are weighted appropriately along with other metrics in Sec. 3.3' — the actual weights are never given.
  • Evolutionary algorithm hyperparameters = not stated
    Table 1 defines g, n, p, Pm, Pc, s, kl, ku but no values are reported; subset sizes and generation counts are unknown.
  • Clusterability k and heatmap pooling factor = k unspecified; pooling factor 8
    k for k-means in Eq. 1 is never given; spatial heatmap pooling by factor 8 (Sec. 3.1) is a hand choice.
assumptions (6)
  • domain assumption 10-epoch YOLOv11n mAP50 on small curated subsets is a faithful ground-truth measure of a synthetic dataset's usefulness for object detection training.
    Sec. 4 (Model Training) uses E=10-epoch training as the target of all regressions; no experiment compares this to converged training or to natural (non-curated) dataset configurations, so the metric's practical validity inherits this assumption.
  • domain assumption Feature-extractor quality is measured by cosine similarity / Euclidean distance between paired real-synthetic image features.
    Sec. 3.2 and Appendix A select GroundingDino-tiny because paired DIMO real/synthetic embeddings have the highest cosine similarity (0.9898); a degenerate encoder mapping all images to one vector would score 1.0, so the criterion does not by itself establish downstream usefulness.
  • domain assumption The three dataset pairs (RarePlanes, DIMO, WASABI) are representative of synthetic-data object detection workloads.
    All three are overhead-imagery or controlled-industrial scenes (Sec. 4); the claim that SDQM 'sets a new standard' for synthetic data evaluation extrapolates from these to the whole task family without a transfer test.
  • domain assumption V-Information computed from a COCO-pretrained YOLOv11n, with class remapping and other predictions ignored, measures object-detection dataset difficulty.
    Sec. 3.1 (Eqs. 2–4): predictive entropy uses COCO class correspondences (e.g., 'airplane' class 5→0) and discards all other predictions; the mapping and discarding are design choices assumed to capture difficulty.
  • standard math Standard nonparametric statistics (K-S, A-D, KL, ED, WD, BD, JS) behave as expected on image-derived distributions.
    Used throughout Sec. 3.1 for pixel, bounding-box, and label distributions; standard background, but finite-sample behavior on these specific quantities is unexamined.
  • domain assumption Clusterability (Eq. 1) measures meaningful real-synthetic embedding overlap.
    LC(R,S) = log of mean squared deviations of per-cluster real proportions from the global proportion; assumes k-means cluster structure is a meaningful axis of domain gap — no external validation.
invented entities (4)
  • SDQM (integrated random-forest metric)
    purpose: Single scalar predicting mAP50 of YOLOv11n trained on a synthetic dataset, without full training
    All evidence is in-paper held-out correlation on the same three datasets used for fitting; no external predictions, no public model artifacts with commit hash, no pre-registered transfer test.
  • V-Information for object detection
    purpose: Measures dataset difficulty/trainability as a sub-metric; strongest individual correlate of mAP50 (Fig. 1)
    Extension of Ethayarajh et al.'s V-usable information to detection via frozen-backbone 10-epoch YOLOv11n fine-tuning; validated only on the paper's in-sample subset pairs.
  • Dataset Separability (AutoKeras trainability)
    purpose: Quantifies how easily real vs synthetic embeddings can be classified, with parameter count correcting for model capacity
    Newly proposed for this paper; correlated in-sample with mAP50 but the weighting of accuracy vs parameters is never specified.
  • Clusterability log metric (Eq. 1)
    purpose: Measures embedding-space overlap of real and synthetic data via k-means
    New formula; author-added. It was later removed from final SDQM by feature selection, and k is unspecified.

how reviews work

0 comments
Cite this review

Pith. "Pith review of SDQM: Synthetic Data Quality Metric for Object Detection Dataset Evaluation." pith.science (2026). https://pith.science/paper/XD64DWWD

@misc{pith2026251006596,
  author       = {Pith},
  title        = {Pith review of: SDQM: Synthetic Data Quality Metric for Object Detection Dataset Evaluation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XD64DWWD}},
  note         = {Machine review of arXiv:2510.06596}
}
read the original abstract

The performance of machine learning models depends heavily on training data. The scarcity of large-scale, well-annotated datasets poses significant challenges in creating robust models. To address this, synthetic data generated through simulations and generative models has emerged as a promising solution, enhancing dataset diversity and improving the performance, reliability, and resilience of models. However, evaluating the quality of this generated data requires an effective metric. We introduce the Synthetic Dataset Quality Metric (SDQM) to assess data quality for object detection tasks without requiring model training to converge. This metric enables more efficient generation and selection of synthetic datasets, addressing a key challenge in resource-constrained object detection tasks. In our experiments, SDQM demonstrated a strong correlation with the mean average precision (mAP) scores of YOLO11, a leading object detection model, whereas previous metrics only exhibited moderate or weak correlations. In addition, it provides actionable insights into improving dataset quality, minimizing the need for costly iterative training. This scalable and efficient metric sets a new standard for evaluating synthetic data. The code for SDQM is available at https://github.com/ayushzenith/SDQM

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

30 extracted references · 4 canonical work pages

  1. [1]

    RarePlanes: Synthetic Data Takes Flight

    J. Shermeyer, T. Hossler, A. Van Etten, et al. , ``Rareplanes: Synthetic data takes flight,'' in Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision , 207--217 (2021). [doi:10.48550/arXiv.2006.02963]

    work page Pith review arXiv doi:10.48550/arxiv.2006.02963 2021

  2. [2]

    P. D. Roovere, S. Moonen, N. Michiels, et al. , ``Dataset of industrial metal objects,'' CoRR abs/2208.04052 (2022). [doi:10.48550/arXiv.2208.04052]

    work page Pith review arXiv doi:10.48550/arxiv.2208.04052 2022

  3. [3]

    M. A. Esposito, J. Lin, R. Young, et al. , ``Odusi: object detection using synthetic imagery,'' in Synthetic Data for Artificial Intelligence and Machine Learning: Tools, Techniques, and Applications II , 13035 , 322--332, SPIE (2024). [doi:10.1117/12.3013516]

    work page doi:10.1117/12.3013516 2024

  4. [4]

    Z. Wang, A. C. Bovik, H. R. Sheikh, et al. , ``Image quality assessment: from error visibility to structural similarity,'' IEEE Trans. Image Process. 13 (4), 600--612 (2004). [doi:10.1109/TIP.2003.819861]

    arXiv 2004

  5. [5]

    Zhang, P

    R. Zhang, P. Isola, A. A. Efros, et al. , ``The unreasonable effectiveness of deep features as a perceptual metric,'' in 2018 IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2018, Salt Lake City, UT, USA, June 18-22, 2018 , 586--595, Computer Vision Foundation / IEEE Computer Society (2018). [doi:10.1109/CVPR.2018.00068]

    arXiv 2018

  6. [6]

    Salimans, I

    T. Salimans, I. J. Goodfellow, W. Zaremba, et al. , ``Improved techniques for training gans,'' in Advances in Neural Information Processing Systems 29: Annual Conference on Neural Information Processing Systems 2016, December 5-10, 2016, Barcelona, Spain , D. D. Lee, M. Sugiyama, U. von Luxburg, et al. , Eds., 2226--2234 (2016)

    2016

  7. [7]

    Heusel, H

    M. Heusel, H. Ramsauer, T. Unterthiner, et al. , ``Gans trained by a two time-scale update rule converge to a local nash equilibrium,'' in Advances in Neural Information Processing Systems 30: Annual Conference on Neural Information Processing Systems 2017, December 4-9, 2017, Long Beach, CA, USA , I. Guyon, U. von Luxburg, S. Bengio, et al. , Eds., 6626-...

    2017

  8. [8]

    Borji, ``Pros and cons of GAN evaluation measures: New developments,'' Comput

    A. Borji, ``Pros and cons of GAN evaluation measures: New developments,'' Comput. Vis. Image Underst. 215 , 103329 (2022). [doi:10.1016/j.cviu.2021.103329]

    arXiv 2022

Show all 30 references
  1. [9]

    Pillutla, L

    K. Pillutla, L. Liu, J. Thickstun, et al. , `` MAUVE scores for generative models: Theory and practice,'' J. Mach. Learn. Res. 24 , 356:1--356:92 (2023)

    2023

  2. [10]

    Kynk \" a \" a nniemi, T

    T. Kynk \" a \" a nniemi, T. Karras, S. Laine, et al. , ``Improved precision and recall metric for assessing generative models,'' in Advances in Neural Information Processing Systems 32: Annual Conference on Neural Information Processing Systems 2019, NeurIPS 2019, December 8-...

    2019

  3. [11]

    A. M. Alaa, B. van Breugel, E. S. Saveliev, et al. , ``How faithful is your synthetic data? sample-level metrics for evaluating and auditing generative models,'' in International Conference on Machine Learning, ICML 2022, 17-23 July 2022, Baltimore, Maryland, USA , K. Chaudhur...

    2022

  4. [12]

    Ethayarajh, Y

    K. Ethayarajh, Y. Choi, and S. Swayamdipta, ``Understanding dataset difficulty with V-usable information,'' in International Conference on Machine Learning, ICML 2022, 17-23 July 2022, Baltimore, Maryland, USA , K. Chaudhuri, S. Jegelka, L. Song, et al. , Eds., Proceedings of ...

    2022

  5. [13]

    Swayamdipta, R

    S. Swayamdipta, R. Schwartz, N. Lourie, et al. , ``Dataset cartography: Mapping and diagnosing datasets with training dynamics,'' in Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing, EMNLP 2020, Online, November 16-20, 2020 , B. Webber, T....

    2020 doi

  6. [14]

    Djolonga, M

    J. Djolonga, M. Lucic, M. Cuturi, et al. , ``Precision-recall curves using information divergence frontiers,'' in The 23rd International Conference on Artificial Intelligence and Statistics, AISTATS 2020, 26-28 August 2020, Online [Palermo, Sicily, Italy] , S. Chiappa and R. C...

    2020

  7. [15]

    H. Jin, F. Chollet, Q. Song, et al. , ``Autokeras: An automl library for deep learning,'' Journal of Machine Learning Research 24 (6), 1--6 (2023)

    2023

  8. [16]

    Fasano and A

    G. Fasano and A. Franceschini, ``A multidimensional version of the kolmogorov--smirnov test,'' Monthly Notices of the Royal Astronomical Society 225 (1), 155--170 (1987). [doi:10.1093/mnras/225.1.155]

    1987 doi

  9. [17]

    Arshad, M

    M. Arshad, M. Rasool, and M. Ahmad, ``Anderson darling and modified anderson darling tests for,'' Pakistan Journal of Applied Sciences 3 (2), 85--88 (2003). [doi:10.3923/jas.2003.85.88]

    2003 doi

  10. [18]

    Kullback and R

    S. Kullback and R. A. Leibler, ``On information and sufficiency,'' The annals of mathematical statistics 22 (1), 79--86 (1951). [doi:10.1214/aoms/1177729694]

    1951

  11. [19]

    M. L. Rizzo and G. J. Sz \'e kely, ``Energy distance,'' wiley interdisciplinary reviews: Computational statistics 8 (1), 27--38 (2016). [doi:10.1002/wics.1375]

    2016 doi

  12. [20]

    Villani et al

    C. Villani et al. , Optimal transport: old and new , vol. 338, Springer (2009). [doi:10.1007/978-3-540-71050-9]

    2009 doi

  13. [21]

    A. Bhattacharyya, ``On a measure of divergence between two statistical populations defined by their probability distribution,'' Bulletin of the Calcutta Mathematical Society 35 , 99--110 (1943)

    1943

  14. [22]

    Lin, ``Divergence measures based on the shannon entropy,'' IEEE Transactions on Information theory 37 (1), 145--151 (1991)

    J. Lin, ``Divergence measures based on the shannon entropy,'' IEEE Transactions on Information theory 37 (1), 145--151 (1991). [doi:10.1109/18.61115]

    1991 doi

  15. [23]

    Jocher and J

    G. Jocher and J. Qiu, ``Ultralytics yolo11,'' (2024)

    2024

  16. [24]

    T. Lin, M. Maire, S. J. Belongie, et al. , ``Microsoft COCO: common objects in context,'' in Computer Vision - ECCV 2014 - 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part V , D. J. Fleet, T. Pajdla, B. Schiele, et al. , Eds., Lecture Note...

    2014 doi

  17. [25]

    Oquab, T

    M. Oquab, T. Darcet, T. Moutakanni, et al. , ``Dinov2: Learning robust visual features without supervision,'' Trans. Mach. Learn. Res. 2024 (2024)

    2024

  18. [26]

    S. Liu, Z. Zeng, T. Ren, et al. , ``Grounding DINO: marrying DINO with grounded pre-training for open-set object detection,'' CoRR abs/2303.05499 (2023). [doi:10.48550/arXiv.2303.05499]

    2023 doi arXiv

  19. [27]

    Radford, J

    A. Radford, J. W. Kim, C. Hallacy, et al. , ``Learning transferable visual models from natural language supervision,'' in Proceedings of the 38th International Conference on Machine Learning, ICML 2021, 18-24 July 2021, Virtual Event , M. Meila and T. Zhang, Eds., Proceedings ...

    2021

  20. [28]

    Kohavi and G

    R. Kohavi and G. H. John, ``Wrappers for feature subset selection,'' Artif. Intell. 97 (1-2), 273--324 (1997). [doi:10.1016/S0004-3702(97)00043-X]

    1997 doi

  21. [29]

    adaptation in natural and artificial systems by john h. holland

    F. Hayes - Roth, ``Review of "adaptation in natural and artificial systems by john h. holland", the u. of michigan press, 1975,'' SIGART Newsl. 53 , 15 (1975). [doi:10.1145/1216504.1216510]

    1975

  22. [30]

    write newline

    " write newline "" before.all 'output.state := FUNCTION blank.sep after.quote 'output.state := FUNCTION fin.entry output.state after.quoted.block = 'skip 'add.period if write newline FUNCTION new.block output.state before.all = 'skip output.state after.quote = after.quoted.blo...

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.