pith. sign in

arxiv: 2605.05328 · v1 · submitted 2026-05-06 · 💻 cs.CV · cs.RO

Query2Uncertainty: Robust Uncertainty Quantification and Calibration for 3D Object Detection under Distribution Shift

Till Beemelmanns , Alexey Nekrasov , Stefan Vilceanu , Jonas Steinhaus , Timo Woopen , Bastian Leibe , Lutz Eckstein This is my paper

Pith reviewed 2026-05-08 17:07 UTC · model grok-4.3

classification 💻 cs.CV cs.RO
keywords 3D object detectionuncertainty quantificationdistribution shiftcalibrationDETRdensity estimationautonomous driving
0
0 comments X

The pith

Density of latent object query features recalibrates both classification and bounding-box uncertainties in 3D detectors under distribution shift.

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

Modern 3D object detectors for autonomous driving produce poorly calibrated uncertainty estimates, and the problem worsens when test conditions differ from training data. Existing post-hoc calibration techniques improve accuracy on matched data but cannot adjust when the distribution shifts. This paper couples a post-hoc calibrator to a density model built on the latent object-query features inside DETR-style detectors. Because these queries already encode location and class information in a compact form, their density serves as a direct indicator of how far the current input has moved from the training distribution. The resulting density signal is used to adjust both classification scores and regression uncertainties together. On multi-view camera and LiDAR detectors the combined method yields lower calibration error than standard post-hoc techniques in both normal and shifted settings.

Core claim

By fitting a density estimator on the feature representations of latent object queries, the method jointly recalibrates classification and bounding-box regression uncertainties, allowing the calibration to adapt automatically when the input distribution changes.

What carries the argument

A density estimator fitted on latent object-query features from DETR-style 3D detectors, used as a proxy to detect distribution shift and adjust model confidences for both classification and bounding-box regression.

If this is right

  • The approach improves calibration on both multi-view camera and LiDAR detectors.
  • It jointly recalibrates classification and bounding-box regression uncertainties.
  • Performance gains hold for both in-distribution and distribution-shifted scenarios.
  • The method requires only the existing latent queries already computed by DETR-style detectors.

Where Pith is reading between the lines

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

  • The same density signal might be used to detect when a detector should trigger retraining or request human labels.
  • Extending the density estimator to non-DETR 3D detectors would test whether query-like internal features are necessary for the adaptation effect.
  • Better uncertainty under shift could reduce overconfident false positives in safety-critical planning modules of autonomous vehicles.

Load-bearing premise

The density of latent object-query features provides a reliable, adaptable proxy for distribution shift that can be used to recalibrate uncertainties.

What would settle it

On a new test set containing a clear distribution shift, if the proposed density-aware calibration shows no reduction in expected calibration error or no improvement in uncertainty metrics relative to standard post-hoc methods, the central claim would be falsified.

Figures

Figures reproduced from arXiv: 2605.05328 by Alexey Nekrasov, Bastian Leibe, Jonas Steinhaus, Lutz Eckstein, Stefan Vilceanu, Till Beemelmanns, Timo Woopen.

Figure 1
Figure 1. Figure 1: Our density-aware calibration method improves reliabil view at source ↗
Figure 2
Figure 2. Figure 2: Overview of Query2Uncertainty We feed 3D features (either LiDAR or multi-view camera) as tokens into a standard DETR-style 3D object detector where they interact with object queries z. The refined queries at the last decoder layer are passed through a feature density estimator, that returns zdense a measure how well each individual query aligns with True Positives queries from the train set. This module is… view at source ↗
Figure 3
Figure 3. Figure 3: Uncertainties under distributional shift. We visualize classification entropy, centroid total variance, and query density for increasing brightness severity levels. The progressive drift of the query density histogram indicates that the density estimator effectively captures distributional shifts in the input data. vector z: zdens(z) = log q(z) ′ − µˆlog q(z) ′ σˆlog q(z) ′ , (13) where µˆlog q(z) ′ and σˆ… view at source ↗
Figure 5
Figure 5. Figure 5: Example Miscal. Area (MCAxyz) for PETR. 0.0 0.2 0.4 0.6 0.8 1.0 Expected Coverage Level p 0.0 0.2 0.4 0.6 0.8 1.0 O b s e r v e d C o v e r a g e L e v el (p) MCAx = 1.677 MCAy = 11.655 MCAz = 4.718 X-Translation Y-Translation Z-Translation Perfect Calibration (a) Uncalibrated KL [56] 0.0 0.2 0.4 0.6 0.8 1.0 Expected Coverage Level p 0.0 0.2 0.4 0.6 0.8 1.0 O b s e r v e d C o v e r a g e L e v el (p) MCAx… view at source ↗
Figure 6
Figure 6. Figure 6: Uncertainty Realism Visualization. Empirical squared Mahalanobis Distance distributions (blue) and an ideal distribution (red) for the centroid covariance for a calibrated (a), overconfident (b) and underconfident case (c). A.2. Calibrator Parameter Count We summarize the number of learnable parameters for each classification and regression calibrator in view at source ↗
Figure 7
Figure 7. Figure 7: In-Distribution PETR with KL [56] - Uncalibrated view at source ↗
Figure 8
Figure 8. Figure 8: In-Distribution PETR with KL [56] calibrated with DA-TS for regression and DA-IR for classification. 17 view at source ↗
Figure 9
Figure 9. Figure 9: In-Distribution PETR - DE [26] view at source ↗
Figure 10
Figure 10. Figure 10: In-Distribution PETR - MCD [13]. 18 view at source ↗
Figure 11
Figure 11. Figure 11: Distribution Shift - Snow Level 2 PETR with KL [56] - Uncalibrated view at source ↗
Figure 12
Figure 12. Figure 12: Distribution Shift - Snow Level 2 PETR with KL [56] calibrated with DA-TS for regression and DA-IR for classification. 19 view at source ↗
Figure 13
Figure 13. Figure 13: Distribution Shift - Brightness Level 1 PETR with KL [56] - Uncalibrated view at source ↗
Figure 14
Figure 14. Figure 14: Distribution Shift - Brightness Level 1 PETR with KL [56] calibrated with DA-TS for regression and DA-IR for classification. 20 view at source ↗
read the original abstract

Reliable uncertainty estimation for 3D object detection is critical for deploying safe autonomous systems, yet modern detectors remain poorly calibrated, especially under distribution shifts. Although post-hoc calibration methods address this issue and provide improved calibration for in-distribution tests, they fail to adapt in distribution-shifted scenarios. In this work, we address this issue and introduce a density-aware calibration method that couples post-hoc calibrators with the feature density of latent object queries from DETR-style 3D object detectors. These queries form a compact, location and class-aware feature, ideal for density estimation, allowing our approach to adjust model confidences in distribution-shift scenarios. By fitting a density estimator on these query features, our approach jointly recalibrates both classification and bounding box regression uncertainties. On both a multi-view camera and LiDAR-based detector, our approach consistently outperforms standard post-hoc methods in both in-distribution and distribution-shifted scenarios. Code available https://tillbeemelmanns.github.io/query2uncertainty/ .

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

3 major / 2 minor

Summary. The manuscript introduces Query2Uncertainty, a density-aware calibration method for 3D object detection that fits a density estimator to latent object query features from DETR-style detectors and couples it with existing post-hoc calibrators to jointly recalibrate both classification scores and bounding-box regression uncertainties. The approach aims to improve calibration under distribution shifts by treating query density as a proxy for shift severity. The authors claim consistent outperformance over standard post-hoc methods on both multi-view camera and LiDAR-based detectors in in-distribution and distribution-shifted scenarios, with code made available.

Significance. If the central claim holds, the work could offer a practical, query-structure-aware extension to post-hoc calibration that addresses a key limitation in deploying 3D detectors for autonomous systems. The public code release supports reproducibility. However, the significance is limited by the absence of direct evidence that query density reliably tracks shift-induced miscalibration for joint class/regression recalibration, rather than providing incidental regularization effects.

major comments (3)
  1. [Experiments] The central assumption that the density of DETR-style object queries serves as an effective, adaptable proxy for distribution shift (enabling joint modulation of classification and bbox regression uncertainties) lacks direct validation. No correlation analysis, scatter plots, or controlled ablations are presented showing that lower query density reliably predicts increased prediction error, higher ECE, or greater shift intensity across the tested detectors and scenarios.
  2. [Method] §4 (Method) and the experimental setup provide insufficient detail on the density estimator implementation (type, hyperparameters, fitting procedure on ID queries only) and the precise mechanism by which density scores adjust the post-hoc calibrators for both classification and regression outputs. This makes it impossible to determine whether the reported gains stem from shift-aware adaptation or from other factors.
  3. [Experiments] The experimental results claim consistent outperformance but report no statistical significance tests, variance across runs, or breakdown by shift type/severity. Without these, the cross-detector and cross-scenario superiority cannot be assessed as robust rather than dataset-specific.
minor comments (2)
  1. [Abstract] The abstract and introduction could more explicitly define the distribution shifts tested (e.g., weather, sensor, or domain changes) and the exact calibration metrics (ECE, NLL, etc.) used for both classification and regression.
  2. [Method] Notation for the density estimator and its coupling to the calibrators should be formalized with equations to improve clarity and reproducibility.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive and detailed feedback. We appreciate the identification of areas where additional evidence and clarity would strengthen the manuscript. We address each major comment below and have prepared revisions to incorporate the suggested improvements.

read point-by-point responses

  1. Referee: [Experiments] The central assumption that the density of DETR-style object queries serves as an effective, adaptable proxy for distribution shift (enabling joint modulation of classification and bbox regression uncertainties) lacks direct validation. No correlation analysis, scatter plots, or controlled ablations are presented showing that lower query density reliably predicts increased prediction error, higher ECE, or greater shift intensity across the tested detectors and scenarios.

    Authors: We agree that direct validation of query density as a proxy for shift severity would provide stronger support for the central assumption. In the revised manuscript, we have added a dedicated analysis subsection containing: (i) Pearson and Spearman correlation coefficients between query density and both prediction errors and ECE, (ii) scatter plots illustrating the relationship across shift intensities for each detector, and (iii) controlled ablations that isolate the contribution of density-based modulation. These results confirm that lower densities reliably correspond to higher errors and miscalibration, thereby validating the proxy role rather than incidental regularization. revision: yes

  2. Referee: [Method] §4 (Method) and the experimental setup provide insufficient detail on the density estimator implementation (type, hyperparameters, fitting procedure on ID queries only) and the precise mechanism by which density scores adjust the post-hoc calibrators for both classification and regression outputs. This makes it impossible to determine whether the reported gains stem from shift-aware adaptation or from other factors.

    Authors: We thank the referee for highlighting the need for greater implementation transparency. Section 4 has been expanded with: the exact density estimator (kernel density estimation using a Gaussian kernel), all hyperparameters and their selection via cross-validation on ID data, the fitting procedure performed exclusively on in-distribution object queries, and the modulation mechanism—normalized density scores are used to adaptively scale the temperature parameter for classification and the variance scaling factor for regression within the post-hoc calibrators. These details clarify that performance gains arise from the shift-aware density modulation. revision: yes

  3. Referee: [Experiments] The experimental results claim consistent outperformance but report no statistical significance tests, variance across runs, or breakdown by shift type/severity. Without these, the cross-detector and cross-scenario superiority cannot be assessed as robust rather than dataset-specific.

    Authors: We acknowledge that statistical rigor and variability reporting are essential for assessing robustness. The revised experiments now include: standard deviations computed over five independent runs with different random seeds, paired t-tests with reported p-values to establish statistical significance of improvements over baselines, and performance breakdowns stratified by shift type (e.g., weather, sensor degradation, scene layout) and severity levels. These additions demonstrate that the outperformance is consistent and statistically supported across detectors and scenarios. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical method with independent experimental validation

full rationale

The paper proposes an empirical calibration technique that fits a density estimator to latent object-query features from DETR-style detectors and couples the resulting density signal to existing post-hoc calibrators. No derivation chain is presented that reduces a claimed prediction or first-principles result to its own fitted inputs by construction. Performance claims rest on comparative experiments across in-distribution and shifted scenarios rather than on any self-referential definition or self-citation load-bearing step. The central modeling choice (query density as a proxy) is an ansatz justified by empirical results, not by a tautological reduction.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption that object queries provide a suitable representation for density-based adaptation to shifts. No free parameters or invented entities are explicitly detailed in the abstract, though density estimator hyperparameters are implicitly required.

free parameters (1)
  • Density estimator hyperparameters
    Parameters such as bandwidth or kernel choice for fitting the density model on query features are likely tuned on training data.
axioms (1)
  • domain assumption Latent object queries from DETR-style 3D detectors form a compact, location- and class-aware feature representation ideal for density estimation.
    Directly stated in the abstract as the basis for coupling with post-hoc calibrators.

pith-pipeline@v0.9.0 · 5500 in / 1299 out tokens · 31151 ms · 2026-05-08T17:07:42.617361+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

58 extracted references · 58 canonical work pages

  1. [1]

    MultiCorrupt: A Multi-Modal Robustness Dataset and Benchmark of LiDAR-Camera Fusion for 3D Object Detection

    Till Beemelmanns, Quan Zhang, Christian Geller, and Lutz Eckstein. MultiCorrupt: A Multi-Modal Robustness Dataset and Benchmark of LiDAR-Camera Fusion for 3D Object Detection. InIEEE Intell. Vehicles Symp., 2024. 6, 8

    work page 2024

  2. [2]

    nuScenes: A Multimodal Dataset for Autonomous Driving

    Holger Caesar, Varun Bankiti, Alex H Lang, Sourabh V ora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Gian- carlo Baldan, and Oscar Beijbom. nuScenes: A Multimodal Dataset for Autonomous Driving. InCVPR, 2020. 2, 3, 4, 6, 11

    work page 2020

  3. [3]

    End- to-end object detection with transformers

    Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End- to-end object detection with transformers. InECCV, pages 213–229. Springer, 2020. 2

    work page 2020

  4. [4]

    Gaussian yolov3: An accurate and fast object detector using localization uncertainty for autonomous driving

    Jiwoong Choi, Dayoung Chun, Hyun Kim, and Hyuk-Jae Lee. Gaussian yolov3: An accurate and fast object detector using localization uncertainty for autonomous driving. InICCV, pages 502–511, 2019. 2

    work page 2019

  5. [5]

    MMDetection3D: OpenMM- Lab next-generation platform for general 3D object detection,

    MMDetection3D Contributors. MMDetection3D: OpenMM- Lab next-generation platform for general 3D object detection,

    work page

  6. [6]

    Can we trust you? on calibration of a probabilistic object detector for autonomous driving

    Di Feng, Lars Rosenbaum, Claudius Glaeser, Fabian Timm, and Klaus Dietmayer. Can we trust you? on calibration of a probabilistic object detector for autonomous driving. InIROS,

    work page

  7. [7]

    2, 6, 7, 8, 13, 15, 16

    work page

  8. [8]

    Waslander, and Klaus Di- etmayer

    Di Feng, Ali Harakeh, Steven L. Waslander, and Klaus Di- etmayer. A review and comparative study on probabilistic object detection in autonomous driving.IEEE Int. Conf. Intell. Transp. Syst., 23(8):9961–9980, 2022. 2

    work page 2022

  9. [9]

    Den- sity estimation using real NVP

    Laurent Dinh, Jascha Sohl-Dickstein, and Samy Bengio. Den- sity estimation using real NVP. InICLR, 2017. 4, 8

    work page 2017

  10. [10]

    Spatialdetr: Robust scalable transformer-based 3d object detection from multi-view camera images with global cross-sensor attention

    Simon Doll, Richard Schulz, Lukas Schneider, Viviane Ben- zin, Markus Enzweiler, and Hendrik PA Lensch. Spatialdetr: Robust scalable transformer-based 3d object detection from multi-view camera images with global cross-sensor attention. InECCV, pages 230–245. Springer, 2022. 2

    work page 2022

  11. [11]

    VOS: Learning What You Don’t Know by Virtual Outlier Synthesis

    Xuefeng Du, Zhaoning Wang, Mu Cai, and Yixuan Li. VOS: Learning What You Don’t Know by Virtual Outlier Synthesis. InICLR, 2022. 2

    work page 2022

  12. [12]

    Li3detr: A lidar based 3d detection transformer

    Gopi Krishna Erabati and Helder Araujo. Li3detr: A lidar based 3d detection transformer. InWACV, pages 4250–4259,

    work page

  13. [13]

    Towards safe autonomous driving: Capture uncertainty in the deep neural network for lidar 3d vehicle detection

    Di Feng, Lars Rosenbaum, and Klaus Dietmayer. Towards safe autonomous driving: Capture uncertainty in the deep neural network for lidar 3d vehicle detection. InIEEE Int. Conf. Intell. Transp. Syst., pages 3266–3273, 2018. 2

    work page 2018

  14. [14]

    Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning

    Yarin Gal and Zoubin Ghahramani. Dropout as a Bayesian Approximation: Representing Model Uncertainty in Deep Learning. InInt. Conf. Mach. Learn., 2017. 2, 6, 7, 12, 13, 14, 18

    work page 2017

  15. [15]

    Concrete dropout,

    Yarin Gal, Jiri Hron, and Alex Kendall. Concrete dropout,

    work page

  16. [16]

    Weinberger

    Chuan Guo, Geoff Pleiss, Yu Sun, and Kilian Q. Weinberger. On calibration of modern neural networks. InInt. Conf. Mach. Learn., page 1321–1330, 2017. 1, 2, 3, 6, 7, 8, 12, 13, 16

    work page 2017

  17. [17]

    Waslander

    Ali Harakeh and Steven L. Waslander. Estimating and Evalu- ating Regression Predictive Uncertainty in Deep Object De- tectors, 2021. 2, 6, 7, 8, 13

    work page 2021

  18. [18]

    Waslander

    Ali Harakeh, Michael Smart, and Steven L. Waslander. BayesOD: A Bayesian Approach for Uncertainty Estimation in Deep Object Detectors. InICRA, pages 87–93. IEEE, 2020. 2

    work page 2020

  19. [19]

    Bounding box regression with uncertainty for accurate object detection

    Yihui He, Chenchen Zhu, Jianren Wang, Marios Savvides, and Xiangyu Zhang. Bounding box regression with uncertainty for accurate object detection. InCVPR, 2018. 2

    work page 2018

  20. [20]

    OCCUQ: Exploring Efficient Uncertainty Quantification for 3D Occupancy Prediction

    Severin Heidrich, Till Beemelmanns, Alexey Nekrasov, Bas- tian Leibe, and Lutz Eckstein. OCCUQ: Exploring Efficient Uncertainty Quantification for 3D Occupancy Prediction. In ICRA, 2025. 2

    work page 2025

  21. [21]

    Practical confidence and prediction intervals

    Tom Heskes. Practical confidence and prediction intervals. In NeurIPS. MIT Press, 1996. 2

    work page 1996

  22. [22]

    What uncertainties do we need in bayesian deep learning for computer vision?NeurIPS, 30,

    Alex Kendall and Yarin Gal. What uncertainties do we need in bayesian deep learning for computer vision?NeurIPS, 30,

    work page

  23. [23]

    Calib3D: Calibrating Model Preferences for Reliable 3D Scene Understanding

    Lingdong Kong, Xiang Xu, Jun Cen, Wenwei Zhang, Liang Pan, Kai Chen, and Ziwei Liu. Calib3D: Calibrating Model Preferences for Reliable 3D Scene Understanding. InWACV, pages 1965–1978, 2025. 2

    work page 1965

  24. [24]

    Accurate uncertainties for deep learning using calibrated re- gression

    V olodymyr Kuleshov, Nathan Fenner, and Stefano Ermon. Accurate uncertainties for deep learning using calibrated re- gression. InICML, pages 2796–2804. PMLR, 2018. 2, 4

    work page 2018

  25. [25]

    Selim Kuzucu, Kemal Oksuz, Jonathan Sadeghi, and Puneet K. Dokania. On calibration of object detectors: Pit- falls, evaluation and baselines. InECCV, 2024. 1, 2, 3, 6, 11

    work page 2024

  26. [26]

    Multivariate confidence calibration for object detection

    Fabian K¨uppers, Jan Kronenberger, Amirhossein Shantia, and Anselm Haselhoff. Multivariate confidence calibration for object detection. InCVPRW, 2020. 2, 3

    work page 2020

  27. [27]

    Simple and Scalable Predictive Uncertainty Estima- tion using Deep Ensembles

    Balaji Lakshminarayanan, Alexander Pritzel, and Charles Blundell. Simple and Scalable Predictive Uncertainty Estima- tion using Deep Ensembles. InNeurIPS, 2017. 2, 6, 7, 12, 13, 14, 18

    work page 2017

  28. [28]

    A Simple Unified Framework for Detecting Out-of-Distribution Samples and Adversarial Attacks

    Kimin Lee, Kibok Lee, Honglak Lee, and Jinwoo Shin. A Simple Unified Framework for Detecting Out-of-Distribution Samples and Adversarial Attacks. InNeurIPS, 2018. 2

    work page 2018

  29. [29]

    Bevformer: Learning bird’s-eye-view representation from multi-camera images via spatiotemporal transformers

    Zhiqi Li, Wenhai Wang, Hongyang Li, Enze Xie, Chonghao Sima, Tong Lu, Yu Qiao, and Jifeng Dai. Bevformer: Learning bird’s-eye-view representation from multi-camera images via spatiotemporal transformers. InECCV. Springer, 2022. 1, 2

    work page 2022

  30. [30]

    GMMSeg: Gaussian Mixture based Generative Semantic Seg- mentation Models

    Chen Liang, Wenguan Wang, Jiaxu Miao, and Yi Yang. GMMSeg: Gaussian Mixture based Generative Semantic Seg- mentation Models. InNeurIPS, 2022. 2 9

    work page 2022

  31. [31]

    PETR: Position Embedding Transformation for Multi-View 3D Object Detection

    Yingfei Liu, Tiancai Wang, Xiangyu Zhang, and Jian Sun. PETR: Position Embedding Transformation for Multi-View 3D Object Detection. InECCV, pages 531–548, 2022. 1, 2, 3, 6

    work page 2022

  32. [32]

    Petrv2: A unified framework for 3d perception from multi-camera images

    Yingfei Liu, Junjie Yan, Fan Jia, Shuailin Li, Qi Gao, Tian- cai Wang, Xiangyu Zhang, and Jian Sun. Petrv2: A unified framework for 3d perception from multi-camera images. In ICCV, 2023. 1, 2

    work page 2023

  33. [33]

    Seed: A simple and effective 3d detr in point clouds

    Zhe Liu, Jinghua Hou, Xiaoqing Ye, Tong Wang, Jingdong Wang, and Xiang Bai. Seed: A simple and effective 3d detr in point clouds. InECCV, 2024. 2

    work page 2024

  34. [34]

    Robust collaborative 3d object detection in presence of pose errors

    Yifan Lu, Quanhao Li, Baoan Liu, Mehrdad Dianati, Chen Feng, Siheng Chen, and Yanfeng Wang. Robust collaborative 3d object detection in presence of pose errors. InICRA, pages 4812–4818. IEEE, 2023. 1, 2, 4

    work page 2023

  35. [35]

    Deep Deterministic Uncer- tainty: A New Simple Baseline

    Jishnu Mukhoti, Andreas Kirsch, Joost van Amersfoort, Philip HS Torr, and Yarin Gal. Deep Deterministic Uncer- tainty: A New Simple Baseline. InCVPR, 2023. 8

    work page 2023

  36. [36]

    Towards improving calibration in ob- ject detection under domain shift.NeurIPS, 35:38706–38718,

    Muhammad Akhtar Munir, Muhammad Haris Khan, M Sar- fraz, and Mohsen Ali. Towards improving calibration in ob- ject detection under domain shift.NeurIPS, 35:38706–38718,

    work page

  37. [37]

    Cal-DETR: Calibrated Detection Transformer.NeurIPS, 2023

    Muhammad Akhtar Munir, Salman Khan, Muhammad Haris Khan, Mohsen Ali, and Fahad Khan. Cal-DETR: Calibrated Detection Transformer.NeurIPS, 2023. 2, 3, 6, 7, 12, 13

    work page 2023

  38. [38]

    D. A. Nix and A. S. Weigend. Estimating the mean and variance of the target probability distribution. In1994 IEEE International Conference on Neural Networks, 1994. IEEE World Congress on Computational Intelligence, pages 55–60 vol.1. IEEE / Institute of Electrical and Electronics Engineers Incorporated, 1994. 2

    work page 1994

  39. [39]

    Kemal Oksuz, Tom Joy, and Puneet K. Dokania. Towards building self-aware object detectors via reliable uncertainty quantification and calibration. InCVPR, 2023. 3

    work page 2023

  40. [40]

    Cooper, and Milos Hauskrecht

    Mahdi Pakdaman Naeini, Gregory F. Cooper, and Milos Hauskrecht. Obtaining well calibrated probabilities using bayesian binning. InAAAI, pages 2901–2907, 2015. 2

    work page 2015

  41. [41]

    Multiclass confidence and localization cali- bration for object detection

    Bimsara Pathiraja, Malitha Gunawardhana, and Muham- mad Haris Khan. Multiclass confidence and localization cali- bration for object detection. InCVPR, pages 19734–19743,

    work page

  42. [42]

    LiDAR-MIMO: Efficient Uncertainty Estimation for LiDAR-based 3D Object Detection

    Matthew Pitropov, Chengjie Huang, Vahdat Abdelzad, Krzysztof Czarnecki, and Steven Waslander. LiDAR-MIMO: Efficient Uncertainty Estimation for LiDAR-based 3D Object Detection. InIEEE Intell. Vehicles Symp., pages 813–820,

    work page

  43. [43]

    Probabilistic Outputs for Support Vector Machines and Comparisons to Regularized Likelihood Methods.Adv

    John Platt. Probabilistic Outputs for Support Vector Machines and Comparisons to Regularized Likelihood Methods.Adv. Large Margin Classif., 10, 1999. 2, 6, 7, 8, 12, 13, 16

    work page 1999

  44. [44]

    Sampling-free epistemic un- certainty estimation using approximated variance propagation

    Janis Postels, Francesco Ferroni, Huseyin Coskun, Nassir Navab, and Federico Tombari. Sampling-free epistemic un- certainty estimation using approximated variance propagation. InCVPR, 2019. 2

    work page 2019

  45. [45]

    Approaching Neural Network Uncertainty Real- ism

    Joachim Sicking, Alexander Kister, Matthias Fahrland, Stefan Eickeler, Fabian H¨uger, Stefan R¨uping, Peter Schlicht, and Tim Wirtz. Approaching Neural Network Uncertainty Real- ism. InNeurIPS 2019 Workshop on Machine Learning for Autonomous Driving, Vancouver, Canada, 2019. 2, 4, 12

    work page 2019

  46. [46]

    Differential evolution–a simple and efficient heuristic for global optimization over continuous spaces.Journal of global optimization, 11(4): 341–359, 1997

    Rainer Storn and Kenneth Price. Differential evolution–a simple and efficient heuristic for global optimization over continuous spaces.Journal of global optimization, 11(4): 341–359, 1997. 6

    work page 1997

  47. [47]

    Post-hoc Un- certainty Calibration for Domain Drift Scenarios

    Christian Tomani, Sebastian Gruber, Muhammed Ebrar Er- dem, Daniel Cremers, and Florian Buettner. Post-hoc Un- certainty Calibration for Domain Drift Scenarios. InCVPR, pages 10119–10127, 2021. 2

    work page 2021

  48. [48]

    Beyond In-Domain Scenarios: Robust Density-Aware Calibration

    Christian Tomani, Futa Kai Waseda, Yuesong Shen, and Daniel Cremers. Beyond In-Domain Scenarios: Robust Density-Aware Calibration. InPMLR, pages 34344–34368,

    work page

  49. [49]

    Exploring Object-Centric Temporal Modeling for Efficient Multi-View 3D Object Detection

    Shihao Wang, Yingfei Liu, Tiancai Wang, Ying Li, and Xi- angyu Zhang. Exploring Object-Centric Temporal Modeling for Efficient Multi-View 3D Object Detection. InICCV, 2023. 2

    work page 2023

  50. [50]

    DETR3D: 3D Object Detection from Multi-view Images via 3D-to-2D Queries

    Yue Wang, Vitor Campagnolo Guizilini, Tianyuan Zhang, Yilun Wang, Hang Zhao, and Justin Solomon. DETR3D: 3D Object Detection from Multi-view Images via 3D-to-2D Queries. InCoRL, pages 180–191. PMLR, 2022. 2

    work page 2022

  51. [51]

    3D Multi-Object Tracking: A Baseline and New Evaluation Metrics.IROS, 2020

    Xinshuo Weng, Jianren Wang, David Held, and Kris Kitani. 3D Multi-Object Tracking: A Baseline and New Evaluation Metrics.IROS, 2020. 1

    work page 2020

  52. [52]

    Cross modal transformer via coordinates encoding for 3d object dectection

    Junjie Yan, Yingfei Liu, Jianjian Sun, Fan Jia, Shuailin Li, Tiancai Wang, and Xiangyu Zhang. Cross modal transformer via coordinates encoding for 3d object dectection. InICCV,

    work page

  53. [53]

    Second: Sparsely em- bedded convolutional detection.Sensors, 18(10):3337, 2018

    Yan Yan, Yuxing Mao, and Bo Li. Second: Sparsely em- bedded convolutional detection.Sensors, 18(10):3337, 2018. 3

    work page 2018

  54. [54]

    Uncertainty Estimation by Density Aware Evidential Deep Learning

    Taeseong Yoon and Heeyoung Kim. Uncertainty Estimation by Density Aware Evidential Deep Learning. InInt. Conf. Mach. Learn., 2024. 5

    work page 2024

  55. [55]

    Obtaining calibrated probability estimates from decision trees and naive Bayesian classifiers

    Bianca Zadrozny and Charles Elkan. Obtaining calibrated probability estimates from decision trees and naive Bayesian classifiers. InInt. Conf. Mach. Learn., pages 609–616, San Francisco, CA, USA, 2001. Morgan Kaufmann Publishers Inc. 2

    work page 2001

  56. [56]

    Transforming Classifier Scores into Accurate Multiclass Probability Estimates.Int

    Bianca Zadrozny and Charles Elkan. Transforming Classifier Scores into Accurate Multiclass Probability Estimates.Int. Conf. Knowl. Discov. Data Min., 2002. 2, 6, 7, 8, 12, 13, 16

    work page 2002

  57. [57]

    Uncertainty-aware voxel based 3d object detection and tracking with von-mises loss,

    Yuanxin Zhong, Minghan Zhu, and Huei Peng. Uncertainty- Aware V oxel based 3D Object Detection and Tracking with von-Mises Loss.arXiv preprint arXiv:2011.02553, 2020. 1, 2, 4, 6, 7, 8, 13, 15, 16, 17, 19, 20

    work page arXiv 2011

  58. [58]

    CenterFormer: Center-based Transformer for 3D Object Detection

    Zixiang Zhou, Xiangchen Zhao, Yu Wang, Panqu Wang, and Hassan Foroosh. CenterFormer: Center-based Transformer for 3D Object Detection. InECCV, 2022. 1, 2 10 Query2Uncertainty: Robust Uncertainty Quantification and Calibration for 3D Object Detection under Distribution Shift Supplementary Material A. Appendix Section Appendix Contents A.1 Uncertainty Evalu...

    work page arXiv 2022