pith. sign in

arxiv: 2605.24722 · v1 · pith:MZALJJTHnew · submitted 2026-05-23 · 💻 cs.CV

Calibrating Probabilistic Object Detectors with Annotator Disagreement

Zhi Qin Tan , Owen Addison , Yunpeng Li This is my paper

Pith reviewed 2026-06-30 13:06 UTC · model grok-4.3

classification 💻 cs.CV
keywords object detectionprobabilistic modelscalibrationannotator disagreementuncertainty estimationambiguous objectsmedical images
0
0 comments X

The pith

Probabilistic object detectors can be calibrated to match annotator disagreement distributions without any ground truth annotations.

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

High annotator disagreement on ambiguous objects poses challenges for object detection since no single ground truth exists. The paper develops a method to train and calibrate detectors using multiple annotations directly. It aligns the detector's class confidence and bounding box uncertainty estimates with the empirical distribution from annotators. This is achieved through new calibration error metrics and both integrated training and post-processing adjustments that apply to common detectors like YOLO.

Core claim

We introduce an interpretable calibration framework for probabilistic object detectors that aligns class confidence and bounding box variance to the annotators' annotation distribution using four evaluation metrics for classification and localization errors, along with train-time calibration and a post-hoc calibrator, all without requiring ground truth annotations.

What carries the argument

The calibration framework consisting of four metrics to measure calibration errors in classification and localization, plus train-time and post-hoc calibration procedures.

If this is right

  • The framework generalizes to many probabilistic object detectors including YOLO families and two-stage detectors.
  • It enables detectors to express meaningful predictive uncertainties for ambiguous objects.
  • Empirical results show superior performance on real-world and synthetic datasets of medical and natural images.
  • Calibration can be performed without access to any ground truth.

Where Pith is reading between the lines

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

  • This approach may improve reliability in applications where object boundaries are inherently ambiguous, such as medical diagnosis.
  • Future work could test if the calibrated uncertainties improve downstream decision-making in detection pipelines.
  • Similar calibration ideas might apply to other vision tasks involving multiple annotations.

Load-bearing premise

The distribution of annotations from multiple annotators serves as an unbiased proxy for the true ambiguity in object detection.

What would settle it

Observing that the calibrated detector's uncertainty estimates do not match the spread of annotations from additional independent annotators on the same ambiguous objects would falsify the calibration effectiveness.

Figures

Figures reproduced from arXiv: 2605.24722 by Owen Addison, Yunpeng Li, Zhi Qin Tan.

Figure 1
Figure 1. Figure 1: Top: Visualization of annotations on the DX-20 dataset. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overall framework of the proposed calibration method, where all annotations are first pre-processed to obtain annotation clusters, [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Comparison of the detection results of (a) YOLOX, (b) YOLOX-G, and (c) YOLOX-G [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The reliability diagrams of YOLOX on VOC-MIX. The [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
read the original abstract

High degrees of disagreement among annotators can exist for ambiguous objects, e.g. in medical images, underscoring the challenges of establishing ground truth annotations in object detection tasks. Despite this, all existing object detectors implicitly require access to ground truth annotations for either training or evaluation. The fundamental questions we target are: How can we learn an object detector with multiple annotators' annotations but without objective ground truth annotations due to object ambiguity, and how can we enable the learned detector to express meaningful model predictive uncertainties in detecting ambiguous objects? To answer these questions, we present an interpretable approach to calibrate probabilistic object detectors, where the calibration goal is to align the class confidence and bounding box variance estimates to the annotators' annotation distribution. We introduce an efficient yet effective framework to calibrate probabilistic object detectors by designing four evaluation metrics to measure calibration errors regarding classification and localization, and proposing a train-time calibration and post-hoc calibrator, all without the need to access any ground truth. This framework is generalizable to many existing probabilistic object detectors, such as the YOLO families and two-stage detectors. Empirical results with real-world and synthetic datasets of medical and natural images demonstrate the superior performance of the proposed framework with three popular object detectors.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper claims to introduce an interpretable calibration framework for probabilistic object detectors that aligns class confidence scores and bounding-box variance estimates to the empirical distribution of multiple annotators' annotations, without requiring objective ground truth. It defines four new evaluation metrics for classification and localization calibration error, proposes both a train-time calibration procedure and a post-hoc calibrator, shows generalizability to YOLO-family and two-stage detectors, and reports superior performance on real-world and synthetic medical and natural-image datasets.

Significance. If the metrics and calibration procedures can be shown to produce uncertainties that reflect latent object ambiguity rather than annotation artifacts, the work would meaningfully advance probabilistic detection in domains with high annotator disagreement such as medical imaging. The absence of free parameters in the core construction and the explicit handling of multiple annotations are positive features.

major comments (2)
  1. [§3.2] §3.2 (Calibration Metrics): the four proposed calibration-error metrics are defined by direct comparison of model outputs to the same empirical class frequencies and bounding-box statistics computed from the annotator set that serves as the calibration target. This construction makes the reported error reduction equivalent to measuring how well the model reproduces the annotation distribution, rendering it impossible to distinguish genuine uncertainty calibration from reproduction of annotator-specific biases or label noise.
  2. [§4] §4 (Experiments): the evaluation protocol uses the same annotator-derived histograms for both training the calibrator and computing the four metrics on held-out images. Without an independent proxy for true ambiguity (e.g., expert consensus on a separate test set or downstream task performance), the claimed superiority over baselines may reflect in-distribution fitting rather than improved calibration.
minor comments (2)
  1. Notation for the four metrics is introduced without an explicit summary table; adding one would improve readability.
  2. The abstract states the framework is 'generalizable to many existing probabilistic object detectors' but the experiments only cover three specific models; a brief discussion of the minimal interface required for other detectors would strengthen the claim.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive comments. We address the two major concerns point by point below, clarifying the design choices in the metrics and evaluation protocol.

read point-by-point responses

  1. Referee: [§3.2] §3.2 (Calibration Metrics): the four proposed calibration-error metrics are defined by direct comparison of model outputs to the same empirical class frequencies and bounding-box statistics computed from the annotator set that serves as the calibration target. This construction makes the reported error reduction equivalent to measuring how well the model reproduces the annotation distribution, rendering it impossible to distinguish genuine uncertainty calibration from reproduction of annotator-specific biases or label noise.

    Authors: The metrics are intentionally defined against the empirical annotator distribution because the paper's explicit goal is to calibrate detectors to match observed annotator disagreement in the absence of objective ground truth. In settings such as medical imaging, where ambiguity is inherent, alignment with the distribution of multiple annotations constitutes the appropriate calibration target rather than an unintended artifact. The framework does not claim to recover an unobserved true label; it provides a measurable way to make model uncertainties consistent with annotator variability. We can add a clarifying sentence in §3.2 to emphasize this distinction. revision: partial

  2. Referee: [§4] §4 (Experiments): the evaluation protocol uses the same annotator-derived histograms for both training the calibrator and computing the four metrics on held-out images. Without an independent proxy for true ambiguity (e.g., expert consensus on a separate test set or downstream task performance), the claimed superiority over baselines may reflect in-distribution fitting rather than improved calibration.

    Authors: The protocol trains the calibrator on one partition of images and evaluates on held-out images, each using its own annotator-derived histograms; this avoids using identical data for fitting and scoring. The reported gains are measured relative to uncalibrated baselines and other methods that do not target the annotator distribution. While an external proxy for latent ambiguity would be valuable, the current design directly assesses fidelity to the observable calibration objective. We can expand the discussion in §4 to note the absence of downstream-task validation as a limitation. revision: partial

Circularity Check

0 steps flagged

No circularity: calibration explicitly defined against empirical annotator distribution as target

full rationale

The paper's framework is self-contained because it explicitly sets the calibration target and error metrics to be the empirical distribution of annotator annotations for both classification and localization, without any derivation chain that reduces a claimed prediction or first-principles result back to fitted inputs by construction. No self-citations, uniqueness theorems, or ansatzes are invoked in a load-bearing way; the method introduces new metrics and calibrators whose purpose is defined as matching the provided annotator statistics. This matches the standard setup for surrogate-ground-truth calibration tasks and does not exhibit any of the enumerated circularity patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Abstract-only view yields limited visibility into parameters or assumptions; the core premise that annotator distributions can replace ground truth is treated as a domain assumption.

axioms (1)
  • domain assumption Annotator annotations form a usable distribution for calibration targets in the absence of objective ground truth
    Invoked as the basis for the calibration goal and metrics.
invented entities (1)
  • Four evaluation metrics for classification and localization calibration errors no independent evidence
    purpose: Measure how well detector outputs align with annotator distributions
    Newly designed components of the framework

pith-pipeline@v0.9.1-grok · 5744 in / 1170 out tokens · 35713 ms · 2026-06-30T13:06:34.626164+00:00 · methodology

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

63 extracted references · 3 canonical work pages · 2 internal anchors

  1. [1]

    Nguyen, D

    D. Nguyen, D. B. Nguyen, H. Q. Nguyen, J. Elliott, N. T. Nguyen, P. Culliton, VinBigData chest x-ray abnormalities detection,https://kaggle.com/competitions/vinbigdata-chest-xray-abnormalities-detection, 2020. Kaggle

    2020

  2. [2]

    I. E. Hamamci, S. Er, E. Simsar, A. E. Yuksel, S. Gultekin, S. D. Ozdemir, K. Yang, H. B. Li, S. Pati, B. Stadlinger, et al., DENTEX: An abnormal tooth detection with dental enumeration and diagnosis benchmark for panoramic x-rays, 2021.arXiv:2305.19112

    work page arXiv 2021

  3. [3]

    R. Dibo, A. Galichin, P. Astashev, D. V. Dylov, O. Y. Rogov, DeepLOC: Deep learning-based bone pathology localization and classification in wrist x-ray images, in: Proc. Anal. Images Social Netw. Texts, 2024, pp. 199–211

    2024

  4. [4]

    Everingham, C

    M. Everingham, C. K. I. W. Luc Van Gool, J. Winn, A. Zisserman, The PASCAL Visual Object Classes Challenge 2007 (VOC2007), Int. J. Comput. Vis. (IJCV) 88 (2010) 303–338

    2007

  5. [5]

    T.-Y. Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan, P. Dollar, L. Zitnick, Microsoft COCO: Common objects in context, in: Proc. Eur. Conf. Comput. Vis. (ECCV), Zurich, Switzerland, 2014, pp. 740–755

    2014

  6. [6]

    P. Sun, H. Kretzschmar, X. Dotiwalla, A. Chouard, V. Patnaik, P. Tsui, J. Guo, Y. Zhou, Y. Chai, B. Caine, V. Vasudevan, W. Han, J. Ngiam, H. Zhao, A. Timofeev, S. Ettinger, M. Krivokon, A. Gao, A. Joshi, Y. Zhang, J. Shlens, Z. Chen, D. Anguelov, Scalability in perception for autonomous driving: Waymo open dataset, in: Proc. IEEE Conf. Comput. Vis. Patte...

    2020

  7. [7]

    Quinn, K

    L. Quinn, K. Tryposkiadis, J. Deeks, H. C. De Vet, S. Mallett, L. B. Mokkink, Y. Takwoingi, S. Taylor-Phillips, A. Sitch, Interobserver variability studies in diagnostic imaging: a methodological systematic review, Brit. J. Radiol. 96 (2023) 20220972

    2023

  8. [8]

    Tschirschwitz, F

    D. Tschirschwitz, F. Klemstein, B. Stein, V. Rodehorst, A dataset for analysing complex document layouts in the digital humanities and its evaluation with krippendorff’s alpha, in: Proc. German Conf. Pattern Recog., Konstanz, Germany, 2022, p. 354–374

    2022

  9. [9]

    Y. Hu, S. Meina, Crowd R-CNN: An object detection model utilizing crowdsourced labels, in: Proc. Int. Conf. Vis. Image Sig. Process. (ICVISP), Bangkok, Thailand, 2020, pp. 1–7

    2020

  10. [10]

    K. Le, T. Tran, H. Pham, H. Nguyen Trung, T. Le, H. Q. Nguyen, Learning from multiple expert annotators for enhancing anomaly detection in medical image analysis, IEEE Access 11 (2023) 14105–14114

    2023

  11. [11]

    Z. Q. Tan, O. Isupova, G. Carneiro, X. Zhu, Y. Li, Bayesian detector combination for object detection with crowdsourced annotations, in: Proc. Eur. Conf. Comput. Vis. (ECCV), Milan, Italy, 2024, pp. 329–346

    2024

  12. [12]

    Kuppers, J

    F. Kuppers, J. Kronenberger, A. Shantia, A. Haselhoff, Multivariate Confidence Calibration for Object Detection, in: IEEE Conf. Comput. Vis. Pattern Recog. (CVPR) Worksh., 2020, pp. 1322–1330

    2020

  13. [13]

    Oksuz, T

    K. Oksuz, T. Joy, P. K. Dokania, Towards building self-aware object detectors via reliable uncertainty quantification and calibration, in: Proc. IEEE Conf. Comput. Vis. Pattern Recog. (CVPR), Vancouver, Canada, 2023, pp. 9263–9274

    2023

  14. [14]

    Mehrtash, W

    A. Mehrtash, W. M. Wells, C. M. Tempany, P. Abolmaesumi, T. Kapur, Confidence calibration and predictive uncertainty estimation for deep medical image segmentation, IEEE Trans. Med. Imag. 39 (2020) 3868–3878

    2020

  15. [15]

    X. Chen, X. Wang, K. Zhang, K.-M. Fung, T. C. Thai, K. Moore, R. S. Mannel, H. Liu, B. Zheng, Y. Qiu, Recent advances and clinical applications of deep learning in medical image analysis, Med. Image Anal. 79 (2022) 102444

    2022

  16. [16]

    Gawlikowski, C

    J. Gawlikowski, C. R. N. Tassi, M. Ali, J. Lee, M. Humt, J. Feng, A. Kruspe, R. Triebel, P. Jung, R. Roscher, M. Shahzad, W. Yang, R. Bamler, X. X. Zhu, A survey of uncertainty in deep neural networks, Artif. Intell. Rev. 56 (2023) 1513–1589

    2023

  17. [17]

    D. Feng, L. Rosenbaum, C. Glaeser, F. Timm, K. Dietmayer, Can we trust you? on calibration of a probabilistic object detector for autonomous driving, in: Proc. IEEE/RSJ Int. Conf. Intell. Robots Syst. (IROS), Macao, China, 2019

    2019

  18. [18]

    M. K. Nowak, J. Cyranka, N. Maslany, A. Kostuch, J. Derbisz, M. Komorkiewicz, P. Siwck, M. J. Wojcik, D. Marchewka, P. Skruch, How Much Noise is There in Labels Generated by Humans? A Method to Validate Automatically Generated Bounding Boxes , in: IEEE Conf. Comput. Vis. Pattern Recog. (CVPR) Worksh., Los Alamitos, CA, USA, 2025, pp. 4371–4380

    2025

  19. [19]

    Tschirschwitz, V

    D. Tschirschwitz, V. Rodehorst, Label Convergence: Defining an Upper Performance Bound in Object Recognition Through Contradictory Annotations , in: Proc. IEEE Winter Conf. Appl. Comput. Vis. (WACV), Tucson, AZ, USA, 2025, pp. 6848–6857. 19

    2025

  20. [20]

    D. Zhou, J. Li, J. Li, J. Huang, Q. Nie, Y. Liu, B.-B. Gao, Q. Wang, P.-A. Heng, G. Chen, Distribution-aware calibration for object detection with noisy bounding boxes, in: Proc. Brit. Mach. Vis. Conf. (BMVC), Glasgow, UK, 2024

    2024

  21. [21]

    Murrugarra-Llerena, C

    J. Murrugarra-Llerena, C. R. Jung, Noise-aware evaluation of object detectors, in: Proc. IEEE Winter Conf. Appl. Comput. Vis. (WACV), Tucson, AZ, USA, 2025, pp. 9322–9331

    2025

  22. [22]

    H. Su, J. Deng, L. Fei-Fei, Crowdsourcing annotations for visual object detection, in: AAAI Workshop, Human Computation, Ontario, Canada, 2012, pp. 40–46

    2012

  23. [23]

    C. Guo, G. Pleiss, Y. Sun, K. Q. Weinberger, On calibration of modern neural networks, in: Proc. Int. Conf. Mach. Learn. (ICML), Sydney, Australia, 2017, pp. 1321–1330

    2017

  24. [24]

    Kumar, P

    A. Kumar, P. S. Liang, T. Ma, Verified uncertainty calibration, in: Proc. Adv. Neural Inform. Process. Syst. (NeurIPS), volume 32, Vancouver, Canada, 2019

    2019

  25. [25]

    Nixon, M

    J. Nixon, M. W. Dusenberry, L. Zhang, G. Jerfel, D. Tran, Measuring calibration in deep learning, in: IEEE Conf. Comput. Vis. Pattern Recog. (CVPR) Worksh., Long Beach, CA, USA, 2019, pp. 38–41

    2019

  26. [26]

    Mukhoti, V

    J. Mukhoti, V. Kulharia, A. Sanyal, S. Golodetz, P. H. S. Torr, P. K. Dokania, Calibrating deep neural networks using focal loss, in: Proc. Adv. Neural Inform. Process. Syst. (NeurIPS), 2020, pp. 15288–15299

    2020

  27. [27]

    D.-B. Wang, L. Feng, M.-L. Zhang, Rethinking calibration of deep neural networks: Do not be afraid of overconfidence, in: Proc. Adv. Neural Inform. Process. Syst. (NeurIPS), 2021, pp. 11809–11820

    2021

  28. [28]

    Kuzucu, K

    S. Kuzucu, K. Oksuz, J. Sadeghi, P. K. Dokania, On calibration of object detectors: Pitfalls, evaluation and baselines, in: Proc. Eur. Conf. Comput. Vis. (ECCV), Milan, Italy, 2024

    2024

  29. [29]

    Y.-J. Park, C. Sobolewski, N. Azizan, Uncertainty quantification in detection transformers: Object-level calibration and image-level reliability, 2026.arXiv:2412.01782

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  30. [30]

    M. A. Munir, M. H. Khan, M. Sarfraz, M. Ali, Towards improving calibration in object detection under domain shift, in: Proc. Adv. Neural Inform. Process. Syst. (NeurIPS), New Orleans, LA, USA, 2022, pp. 38706–38718

    2022

  31. [31]

    Pathiraja, M

    B. Pathiraja, M. Gunawardhana, M. H. Khan, Multiclass confidence and localization calibration for object detection, in: Proc. IEEE Conf. Comput. Vis. Pattern Recog. (CVPR), Vancouver, Canada, 2023, pp. 19734–19743

    2023

  32. [32]

    M. A. Munir, S. Khan, M. H. Khan, M. Ali, F. Khan, Cal-DETR: Calibrated detection transformer, in: Proc. Adv. Neural Inform. Process. Syst. (NeurIPS), New Orleans, LA, USA, 2023

    2023

  33. [33]

    M. A. Munir, M. H. Khan, S. Khan, F. S. Khan, Bridging precision and confidence: A train-time loss for calibrating object detection, in: Proc. IEEE Conf. Comput. Vis. Pattern Recog. (CVPR), Vancouver, Canada, 2023, pp. 11474–11483

    2023

  34. [34]

    Popordanoska, A

    T. Popordanoska, A. Tiulpin, M. B. Blaschko, Beyond classification: Definition and density-based estimation of calibration in object detection, in: Proc. IEEE Winter Conf. Appl. Comput. Vis. (WACV), Waikoloa, HI, USA, 2024, pp. 574–583

    2024

  35. [35]

    K. P. Alexandridis, I. Elezi, J. Deng, A. Nguyen, S. Luo, Fractal calibration for long-tailed object detection, in: Proc. IEEE Conf. Comput. Vis. Pattern Recog. (CVPR), Nashville, TN, USA, 2025

    2025

  36. [36]

    D. Hall, F. Dayoub, J. Skinner, H. Zhang, D. Miller, P. Corke, G. Carneiro, A. Angelova, N. Sünderhauf, Probabilistic object detection: Definition and evaluation, in: Proc. IEEE Winter Conf. Appl. Comput. Vis. (WACV), Snowmass Village, CO, USA, 2020, pp. 1020–1029

    2020

  37. [37]

    D. Feng, A. Harakeh, S. L. Waslander, K. Dietmayer, A review and comparative study on probabilistic object detection in autonomous driving, IEEE Trans Intell. Transportation Syst. 23 (2022) 9961–9980

    2022

  38. [38]

    Miller, L

    D. Miller, L. Nicholson, F. Dayoub, N. Sünderhauf, Dropout sampling for robust object detection in open-set conditions, in: Proc. IEEE Int. Conf. Robot. Automat. (ICRA), Brisbane, Australia, 2018, pp. 3243–3249

    2018

  39. [39]

    Miller, F

    D. Miller, F. Dayoub, M. Milford, N. Sünderhauf, Evaluating merging strategies for sampling-based uncertainty techniques in object detection, in: Proc. IEEE Int. Conf. Robot. Automat. (ICRA), Montreal, Canada, 2019, pp. 2348–2354

    2019

  40. [40]

    Z. Lyu, N. Gutierrez, A. Rajguru, W. J. Beksi, Probabilistic object detection via deep ensembles, in: Eur. Conf. Comput. Vis. (ECCV) Workshop, 2020, pp. 67–75

    2020

  41. [41]

    R. M. Neal, Bayesian Learning for Neural Networks, Springer-Verlag, 1996

    1996

  42. [42]

    IEEE Intell

    M.T.Le, F.Diehl, T.Brunner, A.Knoll, Uncertaintyestimationfordeepneuralobjectdetectorsinsafety-criticalapplications, in: Proc. IEEE Intell. Transportation Syst. Conf., Maui, HA, USA, 2018, pp. 3873–3878. 20

    2018

  43. [43]

    J. Choi, D. Chun, H. Kim, H.-J. Lee, Gaussian YOLOv3: An accurate and fast object detector using localization uncertainty for autonomous driving, in: Proc. IEEE Int. Conf. Comput. Vis. (ICCV), Seoul, Korea, 2019, pp. 502–511

    2019

  44. [44]

    Y. He, C. Zhu, J. Wang, M. Savvides, X. Zhang, Bounding box regression with uncertainty for accurate object detection, in: Proc. IEEE Conf. Comput. Vis. Pattern Recog. (CVPR), Long Beach, CA, USA, 2019, pp. 2883–2892

    2019

  45. [45]

    Kraus, K

    F. Kraus, K. Dietmayer, Uncertainty estimation in one-stage object detection, in: Proc. IEEE Intell. Transportation Syst. Conf., Auckland, New Zealand, 2019, pp. 53–60

    2019

  46. [46]

    Harakeh, M

    A. Harakeh, M. Smart, S. L. Waslander, Bayesod: A bayesian approach for uncertainty estimation in deep object detectors, in: Proc. IEEE Int. Conf. Robot. Automat. (ICRA), 2020, pp. 87–93

    2020

  47. [47]

    Y. He, J. Wang, Deep mixture density network for probabilistic object detection, in: Proc. IEEE/RSJ Int. Conf. Intell. Robots Syst. (IROS), 2020, pp. 10550–10555

    2020

  48. [48]

    Harakeh, S

    A. Harakeh, S. L. Waslander, Estimating and evaluating regression predictive uncertainty in deep object detectors, in: Proc. Int. Conf. Learn. Represent. (ICLR), 2021

    2021

  49. [49]

    S. Su, Y. Li, S. He, S. Han, C. Feng, C. Ding, F. Miao, Uncertainty quantification of collaborative detection for self-driving, in: Proc. IEEE Int. Conf. Robot. Automat. (ICRA), London, United Kingdom, 2023, pp. 5588–5594

    2023

  50. [50]

    Ausiello, P

    G. Ausiello, P. Crescenzi, G. Gambosi, V. Kann, A. Marchetti-Spaccamela, M. Protasi, Complexity and Approximation: Combinatorial Optimization Problems and Their Approximability Properties, Springer Publishing Company, 2013

    2013

  51. [51]

    P. C. Mahalanobis, On the generalized distance in statistics, Proceedings of the National Institute of Sciences of India 2 (1936) 49–55

    1936

  52. [52]

    A. B. Tsybakov, Introduction to Nonparametric Estimation, Springer Series in Statistics, Springer, New York, NY, USA, 2009, p. 83

    2009

  53. [53]

    Krishnamoorthy, Normal distribution, in: Handbook of Statistical Distributions with Applications, CRC Press, Boca Raton, FL, USA, 2006, p

    K. Krishnamoorthy, Normal distribution, in: Handbook of Statistical Distributions with Applications, CRC Press, Boca Raton, FL, USA, 2006, p. 151

    2006

  54. [54]

    S. Ren, K. He, R. Girshick, J. Sun, Faster R-CNN: Towards real-time object detection with region proposal networks, in: Proc. Adv. Neural Inform. Process. Syst. (NeurIPS), Montréal, Canada, 2015, pp. 91–99

    2015

  55. [55]

    Z. Ge, S. Liu, F. Wang, Z. Li, J. Sun, YOLOX: Exceeding YOLO series in 2021, 2021.arXiv:2107.08430

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  56. [56]

    C.-Y. Wang, A. Bochkovskiy, H.-Y. M. Liao, YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors, in: Proc. IEEE Conf. Comput. Vis. Pattern Recog. (CVPR), Vancouver, Canada, 2023, pp. 7464–7475

    2023

  57. [57]

    T.-Y. Lin, P. Goyal, R. Girshick, K. He, P. Dollar, Focal loss for dense object detection, in: Proc. IEEE Int. Conf. Comput. Vis. (ICCV), Venice, Italy, 2017

    2017

  58. [58]

    Robertson, F

    T. Robertson, F. T. Wright, R. Dykstra, Order Restricted Statistical Inference, Wiley, 1988

    1988

  59. [59]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, E. Duchesnay, Scikit-learn: Machine learning in Python, J. Mach. Learn. Res. (JMLR) 12 (2011) 2825–2830

    2011

  60. [60]

    Kendall, Y

    A. Kendall, Y. Gal, What uncertainties do we need in bayesian deep learning for computer vision?, in: Proc. Adv. Neural Inform. Process. Syst. (NeurIPS), Long Beach, California, USA, 2017, pp. 5580––5590

    2017

  61. [61]

    Paszke, S

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga, A. Desmaison, A. Kopf, E. Yang, Z. DeVito, M. Raison, A. Tejani, S. Chilamkurthy, B. Steiner, L. Fang, J. Bai, S. Chintala, Pytorch: An imperative style, high-performance deep learning library, in: Proc. Adv. Neural Inform. Process. Syst. (NeurIP...

    2019

  62. [62]

    Lakshminarayanan, A

    B. Lakshminarayanan, A. Pritzel, C. Blundell, Simple and scalable predictive uncertainty estimation using deep ensembles, in: Proc. Adv. Neural Inform. Process. Syst. (NeurIPS), Long Beach, California, USA, 2017, pp. 6405—-6416

    2017

  63. [63]

    Oksuz, B

    K. Oksuz, B. C. Cam, S. Kalkan, E. Akbas, One metric to measure them all: Localisation recall precision (LRP) for evaluating visual detection tasks, IEEE Trans. Pattern Anal. Mach. Intell. (TPAMI) 44 (2022) 9446–9463. 21

    2022