Pith. sign in

REVIEW 4 major objections 7 minor 41 references

HeAL3D: Heuristical-enhanced Active Learning for 3D Object Detection

T0 review · 4 major / 7 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read HeAL ranks LiDAR scenes by rotation-consistency and distance heuristics, reaching fully supervised 3D detection accuracy with 24% of labels.

desk verdict A plausible, architecture-agnostic AL method whose main headline claim is not backed by the reported experiments. read the letter →

arxiv 2505.00507 v2 pith:YYFF7G2F submitted 2025-05-01 cs.CV

Esteban Rivera , Surya Prabhakaran , Markus Lienkamp This is my paper · ORCID
classification cs.CV
keywords activelearning3DobjectdetectionLiDARGaussianmixturemodelKLdivergenceuncertaintyestimationKITTIbenchmarkpointcloud
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

The paper's aim is to make active learning for LiDAR 3D object detection practical: decide which unlabeled scenes should be sent to a human annotator. HeAL scores each scene by how much the detector's output changes under a 180-degree rotation, then weights that change with object distance and point count. The claim is that this score identifies the most informative samples: on KITTI, training with HeAL-selected samples reaches the same 3D mean average precision as a fully supervised model while using only 24% of the labeled data, and after eight selection cycles it outperforms the KECOR, CRB, and random baselines by 3%, 4.9%, and 4.2% mAP respectively. This would matter because LiDAR labeling is expensive and existing methods plateau in the higher-data regime.

What carries the argument

The central device is a classwise Gaussian mixture model (GMM) built from ordinary detector outputs. Each predicted 3D bounding box becomes a Gaussian blob centered at the box, with covariance taken from its dimensions; a detector's whole scene is thus a sum of blobs per class. The score is the KL divergence between the original scene's GMM and the GMM after a 180-degree rotation that is rotated back into alignment, averaged over classes. Distance and point-count corrections multiply the Gaussian covariance so faraway or sparse objects raise the uncertainty. Because the GMM is built from standard outputs, the mechanism is architecture-agnostic.

What would settle it

On a held-out KITTI split, label the 100 scenes HeAL ranks highest and the 100 it ranks lowest, retrain both from the same starting model, and compare mAP gains: if the low-ranked set improves the detector as much as or more than the high-ranked set over several cycles, then rotation consistency is not the source of the improvement.

Watch

Extended reading notes

Core claim

The central discovery is that localization can drive active learning for 3D detectors if the scene's predicted bounding boxes are turned into a classwise Gaussian mixture model and compared, via KL divergence, between the original point cloud and one rotated 180 degrees around the vertical axis. Two heuristic corrections scale the covariance of each Gaussian with the object's distance from the sensor or with the number of LiDAR points inside the box, so that distant and sparse objects contribute more uncertainty. Averaging the classwise KL divergences gives the HeAL score, and the 100 highest-scoring scenes are labeled each round. Ablations show that separating classes into their own probability maps and adding the distance correction produce most of the gain, while the point-count correction alone harms performance.

Load-bearing premise

The method's load-bearing premise is that a 180-degree rotation does not change how hard a scene is to detect, so any shift in the detector's output under rotation is a faithful signal of which samples deserve labeling.

Editorial extensions

If this is right

  • If the 24% result transfers, an autonomous-driving team can train a LiDAR detector to full-data accuracy while annotating roughly one scene in four.
  • The classwise design should put labeling budget on underrepresented classes; HeAL reports clear AP gains for pedestrians and cyclists while the car class stays flat.
  • Because HeAL uses only standard detection outputs, it can be added to existing pipelines without changing the model or training an extra proxy network.
  • The paper's suggested split, KECOR for the low-data regime and HeAL later, implies that a fixed budget is best spent with two acquisition functions rather than one.
  • The ablation result that point-count correction alone hurts means point sparsity should not be used as a standalone uncertainty signal.

Reading between the lines

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

  • If the mechanism is genuinely about detector instability under label-preserving changes, other perturbations, such as small translations, dropped sensor lines, or temporal shifts, should produce the same ranking; this is testable without new labels.
  • The distance weighting should transfer best to datasets with long-range detections; a natural extension is to normalize distance by class-specific distributions instead of a global maximum.
  • The point-count failure suggests occlusion, not sparsity, is the harder failure mode, so an occlusion-aware heuristic may beat both corrections.
  • The 24% matching point depends on the budget's stopping rule; an economic claim about labeling cost would need AP curves at every budget level for several seeds, which the paper provides for KITTI.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. The paper introduces HeAL, an active learning strategy for LiDAR-based 3D object detection. HeAL represents a scene's detections as a Gaussian mixture model, computes a KL-divergence between the original point cloud and a 180-degree rotated version, incorporates class-wise probability maps, and applies distance- and point-density-based covariance corrections. Experiments on KITTI and a nuScenes subset use PV-RCNN within the ACTIVE-3D-DET toolbox, comparing HeAL against Random, Entropy, Monte Carlo, CRB, and KECOR. The abstract claims competitive state-of-the-art mAP and parity with a full-supervised baseline at 24% of the training samples, with ablation studies attributing gains to the class-wise and distance corrections.

Significance. If the claims are substantiated, HeAL offers a practical and architecture-agnostic acquisition function for 3D object detection, with the notable strength of improving rare-class performance and being built from standard detection outputs. The use of a public toolbox, multiple baselines, and three-seed averaged plots are positive aspects that support reproducibility. However, the central value proposition—parity with full supervision at a fraction of the data—is not demonstrated in the reported results, and the ablations are selected and evaluated on the same KITTI benchmark. These issues currently limit the significance of the contribution to a promising but under-verified heuristic.

major comments (4)
  1. [Abstract and Section 4 (Figure 3)] The abstract's claim that HeAL "achieves the same mAP as the full-supervised baseline with only 24% of the samples" is not supported by any reported full-supervised baseline. Figure 3 and the surrounding text compare HeAL only against Random, Entropy, Monte Carlo, CRB, and KECOR; no PV-RCNN trained on the full KITTI training set is plotted or tabulated, and no full-supervised total mAP value is given anywhere. The manuscript must include the full-supervised baseline curve or value, state explicitly the sample budget at which parity occurs, and reconcile the 24% figure with the experimental protocol (8 cycles of 100 samples is 800 samples, about 21.6% of 3,712; Table 1 reports results at 700 samples).
  2. [Section 4.1, Table 1 vs. Section 4] The ablation in Table 1 identifies GMM+Class+Distance as the best variant on KITTI, and Section 4 then reports HeAL's headline improvements over KECOR, CRB, and Random on the same KITTI benchmark. Because the same dataset is used both to select the variant and to report the final comparison, the reported gains may be optimistic due to selection on the evaluation set. Please provide a clear protocol in which the variant is fixed before the final KITTI comparison, or evaluate the selected variant on a separate held-out split, and report the behavior of all ablated variants across active learning cycles.
  3. [Section 3.5, Eqs. (6)–(8)] The distance and point-count corrections are insufficiently specified to be reproducible and their behavior is not fully justified. In Eq. (6), U_d is normalized by the maximum distance within the detections of a single scene; Eq. (7) uses S, described only as "the total number of points across all bounding boxes," without stating whether this is per scene, per class, or global; and Eq. (8) multiplies the covariance by these corrections. The paper should define all symbols, state the grid resolution and epsilon used in the KL computation, and explain why multiplicatively scaling the covariance produces a meaningful uncertainty estimate rather than simply re-weighting the score by scene composition. The current presentation makes it hard to assess how much of HeAL's gain comes from the augmentation-inconsistency signal versus the heuristic corrections.
  4. [Section 3.6.1 and Figure 4] The nuScenes experimental protocol is inconsistent with the rest of the paper. Section 3.6.1 reports the same sample split (3,712 training, 3,769 validation) for nuScenes as for KITTI, and Figure 4 shows an x-axis extending to 8,000 labeled samples, which exceeds the stated training-set size. Please clarify the actual nuScenes subset size, the active learning budget and cycle structure, and ensure the axis and text are consistent. Without this, the claimed 2% improvement over KECOR on nuScenes cannot be interpreted.
minor comments (7)
  1. [Throughout] There are numerous typographical errors and inconsistent spellings, including "arbitratry" (Contributions), "vainilla" (Section 4.1), "missclasification" (Section 4.1), "HeAL" versus "Heal" (Section 4), and "NuScenes" versus "Nuscenes" (captions and text). The manuscript should be carefully proofread.
  2. [Section 3.5, Eq. (8)] Notation is inconsistent: Eqs. (6) and (7) define U_k^d and U_k^p, while Eq. (8) uses U_d and U_p without the subscript k. Please align the notation.
  3. [Section 4, Figure 4] The legend labels "HeAL(Points)" and "HeAL(Distance)" are not defined in the caption or text. Please clarify that these correspond to the variants in Table 1 and describe exactly which corrections are active in each curve.
  4. [Section 4, first paragraph] The sentence "In the low-data regime, only KECOR has the same performance as the Random baseline" is awkward given that KECOR is described as state-of-the-art; please rephrase to state clearly whether KECOR matches or exceeds Random in this regime.
  5. [Section 4, Figure 5] The statement "HeAL outperforms the concurrent method for all difficulty levels" is vague; please name the concurrent method (presumably KECOR) and, if possible, show all baselines per difficulty level rather than only HeAL and one comparison method.
  6. [Section 4.1, Table 1] The table reports mean mAP values but no standard deviations, despite the text stating that all plots show mean results over three seeds. Please include variance information or state why it is omitted for the ablation.
  7. [Section 3.6.3] The initial labeled set size is never specified; Section 3.1 defines D_L with cardinality m, but no value is given for the experiments. This is needed to reproduce the active learning curves.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: HeAL's score is a hand-designed, parameter-free heuristic derived from detector outputs; no reported prediction reduces to its own inputs by construction.

full rationale

The HeAL acquisition score is defined entirely by fixed formulas: Eq. (1)-(2) build a GMM from detector bounding boxes, Eq. (3)-(5) compute KL divergence between original and 180-degree-rotated detections, and Eq. (6)-(8) apply distance and point-count corrections. None of these formulas is fitted to the mAP that is later reported; the score is used only to select samples, and the mAP is measured after retraining. Thus the central derivation is self-contained and not circular. The only self-citation relevant to the method is [27], used alongside [18] to justify the 180-degree rotation augmentation in Section 3.1. This is not load-bearing: the paper independently explains the choice by scene plausibility, and no uniqueness theorem or 'by construction' equivalence is invoked. The most serious issue is the abstract's claim of matching the full-supervised baseline with 24% of the samples; Section 4 and Table 1 report comparisons against Random, Entropy, Monte Carlo, CRB, and KECOR but never report a full-supervised PV-RCNN mAP, and the 24% figure is not tied cleanly to the eight-cycle protocol (800/3712 = 21.6%). That is an omitted-comparison and completeness problem, not circularity. The ablation in Table 1 selects the best heuristic combination on the same KITTI benchmark used for the headline, which introduces mild benchmark-selection risk, but the selected modules are not fitted parameters of an acquisition function that numerically force the final mAP. Overall, there is no demonstrated reduction of the prediction to its inputs, so the circularity score is minimal.

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

The method rests on several unproven modeling choices beyond standard active learning assumptions. The GMM representation, the use of box dimensions as covariance, the assumption that distance and point count signal difficulty, and the classwise KL averaging are all chosen by hand and validated only on the benchmark where they are reported.

free parameters (5)
  • sigma_scale = sqrt(l), sqrt(w), sqrt(h)
    Eq. 1 sets the Gaussian standard deviations to the square root of the box dimensions; a hand-chosen mapping that determines the shape of every localization Gaussian.
  • epsilon_point = 1e-9
    Eq. 7 includes epsilon to avoid division by zero; its value is arbitrary and unchecked.
  • grid_resolution = not reported
    The spatial grid on which Eqs. 1-4 evaluate the Gaussians is not specified; the KL scores depend on it.
  • initial_labeled_set_size = not reported
    The size of the initial labeled set D_L is not reported; the headline 24% claim depends on it.
  • selected_ablation_variant = Localization+Class+Distance
    The best variant was chosen using the KITTI 700-sample ablation, so the reported KITTI result is on a variant selected by the same benchmark.
assumptions (5)
  • domain assumption Rotating a point cloud 180 degrees around the z-axis produces a plausible scene whose detections should be consistent with the original.
    Section 3.1: the consistency between original and augmented predictions is used as uncertainty; if the augmentation changes detection difficulty systematically, the inconsistency score is biased.
  • ad hoc to paper A 3D detection can be represented as a Gaussian centered at the box center with diagonal covariance derived from box dimensions.
    Section 3.2, Eq. 1: the variance is set to the box dimensions, a modeling choice not derived from the detector's actual output distribution.
  • domain assumption Distant and low-point-density objects are harder to detect, so their uncertainty should be weighted higher.
    Section 3.5, Eqs. 6-7: this assumption from the detection literature drives the heuristic corrections.
  • domain assumption Samples whose predictions are inconsistent under a light augmentation are the most informative to label.
    Section 2.4: consistency-based active learning assumes disagreement under augmentation indicates information gain.
  • domain assumption Splitting the GMM by predicted class and averaging KL per class yields a better uncertainty estimate than a single scene-level heatmap.
    Section 3.4: the paper asserts this but the ablation is the only support; no theoretical justification is given.

how reviews work

0 comments
Cite this review

Pith. "Pith review of HeAL3D: Heuristical-enhanced Active Learning for 3D Object Detection." pith.science (2026). https://pith.science/paper/YYFF7G2F

@misc{pith2026250500507,
  author       = {Pith},
  title        = {Pith review of: HeAL3D: Heuristical-enhanced Active Learning for 3D Object Detection},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YYFF7G2F}},
  note         = {Machine review of arXiv:2505.00507}
}
read the original abstract

Active Learning has proved to be a relevant approach to perform sample selection for training models for Autonomous Driving. Particularly, previous works on active learning for 3D object detection have shown that selection of samples in uncontrolled scenarios is challenging. Furthermore, current approaches focus exclusively on the theoretical aspects of the sample selection problem but neglect the practical insights that can be obtained from the extensive literature and application of 3D detection models. In this paper, we introduce HeAL (Heuristical-enhanced Active Learning for 3D Object Detection) which integrates those heuristical features together with Localization and Classification to deliver the most contributing samples to the model's training. In contrast to previous works, our approach integrates heuristical features such as object distance and point-quantity to estimate the uncertainty, which enhance the usefulness of selected samples to train detection models. Our quantitative evaluation on KITTI shows that HeAL presents competitive mAP with respect to the State-of-the-Art, and achieves the same mAP as the full-supervised baseline with only 24% of the samples.

Figures

Figures reproduced from arXiv: 2505.00507 by the authors.

Figure 1
Figure 1. Typical driving scenario for an autonomous vehicle. Even [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. HeAL Score calculation pipeline, presented as BEV for clarity whereas the actual procedure is done on 3D. 1) Our input is a 3D [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. mAP in % with respect to the number of labeled samples [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: mAP in KITTI for each difficulty level [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: AP in KITTI for each class [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 37 canonical work pages

  1. [1]

    Ash, Chicheng Zhang, Akshay Krishnamurthy, John Langford, and Alekh Agarwal

    Jordan T. Ash, Chicheng Zhang, Akshay Krishnamurthy, John Langford, and Alekh Agarwal. Deep batch active learning by diverse, uncertain gradient lower bounds. In ICLR. ACM,

    work page

  2. [2]

    Beluch, Tim Genewein, Andreas Nurnberger, and Jan M

    William H. Beluch, Tim Genewein, Andreas Nurnberger, and Jan M. Kohler. The power of ensembles for active learning in image classification. In 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9368–9377,

    work page 2018

  3. [3]

    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, Gi- ancarlo Baldan, and Oscar Beijbom. nuscenes: A multimodal dataset for autonomous driving. In 2020 IEEE/CVF con- ference on computer vision and pattern recognition , pages 11621–11631, 2020. 1, 6

    work page 2020

  4. [4]

    Bayesian active learning by disagreements: A geometric perspective

    Xiaofeng Cao and Ivor Wai-Hung Tsang. Bayesian active learning by disagreements: A geometric perspective. ArXiv, abs/2105.02543, 2021. 2

    arXiv 2021

  5. [5]

    Se- quential graph convolutional network for active learning

    Razvan Caramalau, Binod Bhattarai, and Tae-Kyun Kim. Se- quential graph convolutional network for active learning. In 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 9578–9587, 2021. 2

    work page 2021

  6. [6]

    Qi Charles, Hao Su, Mo Kaichun, and Leonidas J

    R. Qi Charles, Hao Su, Mo Kaichun, and Leonidas J. Guibas. Pointnet: Deep learning on point sets for 3d classification and segmentation. In 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 77–85, 2017. 6

    work page 2017

  7. [7]

    Jiwoong Choi, Ismail Elezi, Hyuk-Jae Lee, Clement Farabet, and Jose M. Alvarez. Active learning for deep object detection via probabilistic modeling. In 2021 IEEE/CVF International Conference on Computer Vision (ICCV), pages 10244–10253,

    work page 2021

  8. [8]

    Contrastive coding for active learning under class distribution mismatch

    Pan Du, Suyun Zhao, Hui Chen, Shuwen Chai, Hong Chen, and Cuiping Li. Contrastive coding for active learning under class distribution mismatch. In 2021 IEEE/CVF International Conference on Computer Vision (ICCV), pages 8907–8916,

    work page 2021

Show all 41 references
  1. [9]

    Ismail Elezi, Zhiding Yu, Anima Anandkumar, Laura Leal- Taix´e, and Jose M. Alvarez. Not all labels are equal: Ratio- nalizing the labeling costs for training object detection. In 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 14472–14481, 2022. 2, 3

    2022

  2. [10]

    Deep active learning for efficient training of a lidar 3d object detector

    Di Feng, Xiao Wei, Lars Rosenbaum, Atsuto Maki, and Klaus Dietmayer. Deep active learning for efficient training of a lidar 3d object detector. In 2019 IEEE Intelligent Vehicles Symposium (IV), pages 667–674, 2019. 6

    2019

  3. [11]

    Man truckscenes: A multimodal dataset for autonomous trucking in diverse conditions

    Felix Fent, Fabian Kuttenreich, Florian Ruch, Farija Rizwin, Stefan Juergens, Lorenz Lechermann, Christian Nissler, An- drea Perl, Ulrich V oll, Min Yan, and Markus Lienkamp. Man truckscenes: A multimodal dataset for autonomous trucking in diverse conditions. In Advances in Ne...

    2024

  4. [12]

    Dropout as a bayesian approximation: Representing model uncertainty in deep learn- ing

    Yarin Gal and Zoubin Ghahramani. Dropout as a bayesian approximation: Representing model uncertainty in deep learn- ing. In Proceedings of The 33rd International Conference on Machine Learning, pages 1050–1059, New York, New York, USA, 2016. PMLR. 2

    2016

  5. [13]

    Deep bayesian active learning with image data

    Yarin Gal, Riashat Islam, and Zoubin Ghahramani. Deep bayesian active learning with image data. In Proceedings of the 34th International Conference on Machine Learning - Volume 70, page 1183–1192. JMLR.org, 2017. 2

    2017

  6. [14]

    Are we ready for autonomous driving? the kitti vision benchmark suite

    Andreas Geiger, Philip Lenz, and Raquel Urtasun. Are we ready for autonomous driving? the kitti vision benchmark suite. In 2012 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3354–3361. IEEE, 2012. 1, 6

    2012

  7. [15]

    Trivedi, and Alois C

    Ahmed Ghita, Bjørk Antoniussen, Walter Zimmer, Ross Greer, Christian Creß, Andreas Møgelmose, Mohan M. Trivedi, and Alois C. Knoll. Activeanno3d - an active learning framework for multi-modal 3d object detection. In2024 IEEE Intelligent Vehicles Symposium (IV), pages 1699–1706...

    2024

  8. [16]

    Andersen, An- dreas Mogelmose, and Mohan M

    Ross Greer, Bjork Antoniussen, Mathias V . Andersen, An- dreas Mogelmose, and Mohan M. Trivedi. The why, when, and how to use active learning in large-data-driven 3d object detection for safe autonomous driving: An empirical explo- ration. arXiv.org, abs/2401.16634, 2024. 3

    2024 arXiv

  9. [17]

    Deep active learning for biased datasets via fisher kernel self-supervision

    Denis Gudovskiy, Alec Hodgkinson, Takuya Yamaguchi, and Sotaro Tsukizawa. Deep active learning for biased datasets via fisher kernel self-supervision. In 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 9038–9046, 2020. 2

    2020

  10. [18]

    Efficient active learning strategies for monocular 3d object detection

    Aral Hekimoglu, Michael Schmidt, Alvaro Marcos-Ramiro, and Gerhard Rigoll. Efficient active learning strategies for monocular 3d object detection. In 2022 IEEE Intelligent Vehicles Symposium (IV), pages 295–302, 2022. 2, 3, 4

    2022

  11. [19]

    Bayesian active learning for classification and pref- erence learning

    Neil Houlsby, Ferenc Huszar, Zoubin Ghahramani, and M´at´e Lengyel. Bayesian active learning for classification and pref- erence learning. CoRR, abs/1112.5745, 2011. 2

    2011 arXiv

  12. [20]

    Joint semi-supervised and active learning via 3d consistency for 3d object detection

    Sihwan Hwang, Sanmin Kim, Youngseok Kim, and Dong- suk Kum. Joint semi-supervised and active learning via 3d consistency for 3d object detection. In 2023 IEEE Interna- tional Conference on Robotics and Automation (ICRA), pages 4819–4825. IEEE, 2023. 4

    2023

  13. [21]

    Consistency-based semi-supervised learning for object de- tection

    Jisoo Jeong, Seungeui Lee, Jeesoo Kim, and Nojun Kwak. Consistency-based semi-supervised learning for object de- tection. Curran Associates Inc., Red Hook, NY , USA, 2019. 3

    2019

  14. [22]

    Localization-aware active learning for object detection

    Chieh-Chi Kao, Teng-Yok Lee, Pradeep Sen, and Ming-Yu Liu. Localization-aware active learning for object detection. In Computer Vision – ACCV 2018 , pages 506–522, Cham,

    2018

  15. [23]

    Kingma and Jimmy Ba

    Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In 3rd International Conference on Learning Representations, ICLR 2015, San Diego, CA, USA, May 7-9, 2015, Conference Track Proceedings, 2015. 6 9

    2015

  16. [24]

    Kecor: Kernel Coding Rate Maximization for Active 3D Object Detection

    Yadan Luo, Zhuoxiao Chen, Zhen Fang, Zheng Zhang, Mahsa Baktashmotlagh, and Zi Huang. Kecor: Kernel Coding Rate Maximization for Active 3D Object Detection . In 2023 IEEE/CVF International Conference on Computer Vision (ICCV), pages 18233–18244, Los Alamitos, CA, USA, 2023. IE...

    2023

  17. [25]

    Exploring active 3d object de- tection from a generalization perspective

    Yadan Luo, Zhuoxiao Chen, Zijian Wang, Xin Yu, Zi Huang, and Mahsa Baktashmotlagh. Exploring active 3d object de- tection from a generalization perspective. In The Eleventh International Conference on Learning Representations, ICLR 2023, Kigali, Rwanda, May 1-5, 2023. OpenRevi...

    2023

  18. [26]

    Active Learning by Feature Mixing

    Amin Parvaneh, Ehsan Abbasnejad, Damien Teney, Reza Haf- fari, Anton Van Den Hengel, and Javen Qinfeng Shi. Active Learning by Feature Mixing . In 2022 IEEE/CVF Confer- ence on Computer Vision and Pattern Recognition (CVPR), pages 12227–12236, Los Alamitos, CA, USA, 2022. IEEE...

    2022

  19. [27]

    Camera-lidar inconsistency analysis for active learning in object detection

    Esteban Rivera, Ana Clara Serra Do Nascimento, and Markus Lienkamp. Camera-lidar inconsistency analysis for active learning in object detection. In 35th IEEE Intelligent Vehicles Symposium, IV 2024 , pages 97–103. Institute of Electrical and Electronics Engineers Inc., 2024. P...

    2024

  20. [28]

    Ortho- graphic feature transform for monocular 3d object detection

    Thomas Roddick, Alex Kendall, and Roberto Cipolla. Ortho- graphic feature transform for monocular 3d object detection. British Machine Vision Conference, 2019. 1, 4

    2019

  21. [29]

    Active learning for convolu- tional neural networks: A core-set approach

    Ozan Sener and Silvio Savarese. Active learning for convolu- tional neural networks: A core-set approach. In International Conference on Learning Representations, 2018. 2

    2018

  22. [30]

    H. S. Seung, M. Opper, and H. Sompolinsky. Query by committee. In Proceedings of the Fifth Annual Workshop on Computational Learning Theory, page 287–294, New York, NY , USA, 1992. Association for Computing Machinery. 2

    1992

  23. [31]

    Pv-rcnn: Point- voxel feature set abstraction for 3d object detection

    Shaoshuai Shi, Chaoxu Guo, Li Jiang, Zhe Wang, Jianping Shi, Xiaogang Wang, and Hongsheng Li. Pv-rcnn: Point- voxel feature set abstraction for 3d object detection. In Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10529–10538, 2020. 1, 6

    2020

  24. [32]

    Scalability in perception for autonomous driving: Waymo open dataset

    Pei Sun, Henrik Kretzschmar, Xerxes Dotiwalla, Aurelien Chouard, Vijaysai Patnaik, Paul Tsui, James Guo, Yin Zhou, Yuning Chai, Benjamin Caine, Vijay Vasudevan, Wei Han, Jiquan Ngiam, Hang Zhao, Aleksei Timofeev, Scott Et- tinger, Maxim Krivokon, Amy Gao, Aditya Joshi, Yu Zhan...

    2020

  25. [33]

    RSN: Range Sparse Net for Efficient, Accurate LiDAR 3D Object Detection

    Pei Sun, Weiyue Wang, Yuning Chai, Gamaleldin Elsayed, Alex Bewley, Xiao Zhang, Cristian Sminchisescu, and Dragomir Anguelov. RSN: Range Sparse Net for Efficient, Accurate LiDAR 3D Object Detection . In 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)...

    2021

  26. [34]

    Reid, and G

    Toan Tran, Thanh-Toan Do, Ian D. Reid, and G. Carneiro. Bayesian generative active deep learning. In International Conference on Machine Learning, 2019. 2

    2019

  27. [35]

    A new active labeling method for deep learning

    Dan Wang and Yi Shang. A new active labeling method for deep learning. In 2014 International Joint Conference on Neural Networks (IJCNN), pages 112–119, 2014. 6

    2014

  28. [36]

    Basal: Size- balanced warm start active learning for lidar semantic segmen- tation

    Jiarong Wei, Yancong Lin, and Holger Caesar. Basal: Size- balanced warm start active learning for lidar semantic segmen- tation. In 2024 IEEE International Conference on Robotics and Automation (ICRA), pages 18258–18264, 2024. 3

    2024

  29. [37]

    Entropy-based Active Learning for Object Detection with Progressive Diversity Constraint

    Jiaxi Wu, Jiaxin Chen, and Di Huang. Entropy-based Active Learning for Object Detection with Progressive Diversity Constraint . In 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 9387–9396, Los Alamitos, CA, USA, 2022. IEEE Computer Society. 2

    2022

  30. [38]

    Learning Loss for Active Learning

    Donggeun Yoo and In So Kweon. Learning Loss for Active Learning . In 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 93–102, Los Alamitos, CA, USA, 2019. IEEE Computer Society. 2

    2019

  31. [39]

    Consistency-based active learning for object detection

    Weiping Yu, Sijie Zhu, Taojiannan Yang, and Chen Chen. Consistency-based active learning for object detection. In 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pages 3950–3959, 2022. 3

    2022

  32. [40]

    Bi3D: Bi-Domain Ac- tive Learning for Cross-Domain 3D Object Detection

    Jiakang Yuan, Bo Zhang, Xiangchao Yan, Tao Chen, Bo- tian Shi, Yikang Li, and Yu Qiao. Bi3D: Bi-Domain Ac- tive Learning for Cross-Domain 3D Object Detection . In 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 15599–15608, Los Alamitos, CA, U...

    2023

  33. [2019]

    Springer International Publishing. 2

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.