pith. machine review for the scientific record. sign in

arxiv: 2605.10349 · v1 · submitted 2026-05-11 · 💻 cs.CV · cs.AI· cs.LG

Recognition: 2 theorem links

· Lean Theorem

Portable Active Learning for Object Detection

Rashi Sharma , Justin Timothy C. Bersamin , Karthikk Subramanian
Authors on Pith no claims yet

Pith reviewed 2026-05-12 04:47 UTC · model grok-4.3

classification 💻 cs.CV cs.AIcs.LG
keywords active learningobject detectionlabel efficiencyportable frameworkinference outputsuncertainty samplingdata selectionbounding box annotation
0
0 comments X

The pith

Portable Active Learning selects data for object detectors using only inference outputs and no model changes

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

The paper introduces Portable Active Learning (PAL) as a detector-agnostic framework that selects which images to annotate next for object detection tasks. It trains lightweight class-specific logistic classifiers solely on the detector's output predictions to compute uncertainty scores, then refines candidate batches by adding global image entropy, class diversity, and similarity checks. This avoids any access to internal model features or alterations to training pipelines, addressing the high cost of bounding box annotation in real-world settings. Experiments on COCO, PASCAL VOC, and BDD100K show consistent gains in label efficiency and final accuracy over prior active learning baselines. A reader would care because the method promises to make active learning usable across many different detectors without custom integration work.

Core claim

PAL is a detector-agnostic active learning framework that operates solely on inference outputs. At each round it trains lightweight class-specific logistic classifiers to distinguish true from false positives and produce entropy-based uncertainty scores for proposals. Candidate images are then refined using global image entropy, class diversity, and image similarity to form batches that are both informative and diverse. The approach requires no changes to model internals or training pipelines and yields improved label efficiency and detection accuracy on COCO, PASCAL VOC, and BDD100K compared to existing baselines.

What carries the argument

Lightweight class-specific logistic classifiers trained on inference outputs to generate entropy-based uncertainty scores, combined with global image entropy, class diversity, and similarity for batch selection

If this is right

  • PAL integrates with any object detector without modifying its code or training schedule
  • It produces higher detection accuracy for the same annotation budget on COCO, PASCAL VOC, and BDD100K
  • Data selection jointly accounts for instance-level uncertainty, class imbalance, and image-level diversity
  • The framework supports deployment where model internals remain inaccessible

Where Pith is reading between the lines

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

  • The same output-only approach could enable active learning on proprietary detection APIs that expose only predictions
  • Similar portable uncertainty estimation might transfer to related tasks such as instance segmentation
  • The method's robustness could be tested on datasets with extreme class imbalance to check whether the per-class classifiers remain effective

Load-bearing premise

Lightweight class-specific logistic classifiers trained only on inference outputs can reliably separate true positives from false positives to produce useful uncertainty scores

What would settle it

If the logistic classifiers perform no better than random at distinguishing true from false positives on a held-out set and PAL then fails to beat random selection or strong baselines on label efficiency

Figures

Figures reproduced from arXiv: 2605.10349 by Justin Timothy C. Bersamin, Karthikk Subramanian, Rashi Sharma.

Figure 1
Figure 1. Figure 1: Architectural overview of our Portable Active Learning (PAL) framework. The object detector, trained on the current iteration’s [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: PAL’s class-specific logistic classifiers (CLC) across AL data selection rounds (1–4) for the bus category, a low-frequency class [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Feature space of PAL’s logistic classifier visualizing detections across classes on the COCO dataset. Marker color encodes LIUS [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Comparison between PAL and state-of-the-art active learning algorithms. Subfigures (a) and (b) report AP (%) on COCO using [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Ablation study on impact of: (a) different components of GUIDE; (b) GUIDE score contribution to the selection score; (c) [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Classification boundary for XGBoost and simple logistic [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
read the original abstract

Annotating bounding boxes is costly and limits the scalability of object detection. This challenge is compounded by the need to preserve high accuracy while minimizing manual effort in real-world applications. Prior active learning methods often depend on model features or modify detector internals and training schedules, increasing integration overhead. Moreover, they rarely jointly exploit the benefits of image-level signals, class-imbalance cues, and instance-level uncertainty for comprehensive selection. We present Portable Active Learning (PAL), a detector-agnostic, easily portable framework that operates solely on inference outputs. PAL combines class-wise instance uncertainty with image-level diversity to guide data selection. At each round, PAL trains lightweight class-specific logistic classifiers to distinguish true from false positives, producing entropy-based uncertainty scores for proposals. Candidate images are then refined using global image entropy, class diversity, and image similarity, yielding batches that are both informative and diverse. PAL requires no changes to model internals or training pipelines, ensuring broad compatibility across detectors. Extensive experiments on COCO, PASCAL VOC, and BDD100K demonstrate that PAL consistently improves label efficiency and detection accuracy compared to existing active learning baselines, making it a practical solution for scalable and cost-effective deployment of object detection in real-world settings.

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 / 3 minor

Summary. The manuscript proposes Portable Active Learning (PAL), a detector-agnostic framework for active learning in object detection that operates exclusively on inference outputs without requiring access to model internals or changes to training pipelines. PAL trains lightweight class-specific logistic classifiers on proposals (matched to ground-truth labels from the labeled pool) to derive entropy-based uncertainty scores distinguishing true from false positives; these are then combined with image-level signals including global entropy, class diversity, and similarity to select informative and diverse batches. Experiments on COCO, PASCAL VOC, and BDD100K are presented as demonstrating consistent gains in label efficiency and final detection accuracy relative to existing active learning baselines.

Significance. If the empirical claims hold, the work has moderate practical significance for real-world object detection deployments where detectors are treated as black boxes (e.g., commercial APIs or fixed legacy models). By avoiding feature extraction or pipeline modifications, PAL lowers integration barriers compared with many prior active-learning methods that rely on internal activations or retraining schedules. The explicit joint use of instance uncertainty and image-level diversity metrics addresses a noted gap in the literature, and the multi-dataset evaluation provides some evidence of portability.

major comments (2)
  1. [§3] §3 (PAL framework description): the claim that class-specific logistic classifiers trained solely on inference outputs reliably produce useful entropy uncertainty scores rests on an untested assumption about the separability of true/false positives; no ablation is shown on classifier architecture, feature representation (e.g., box coordinates vs. scores), or training-set size, which directly affects the portability and robustness assertions.
  2. [§4] §4 (Experiments): while the abstract and introduction assert 'consistent improvements' across COCO, PASCAL VOC, and BDD100K, the reported results lack explicit quantitative tables or figures showing mAP deltas, label-efficiency curves, and statistical significance against named baselines (e.g., uncertainty sampling, core-set, or diversity-only methods); without these numbers the central empirical claim cannot be evaluated for effect size or reproducibility.
minor comments (3)
  1. [§3.2] Notation for the entropy computation (Eq. 2 or equivalent) should explicitly define the input features to the logistic classifiers and the exact matching procedure between proposals and ground-truth boxes.
  2. [§3.3] The similarity metric used for image-level refinement is described only at a high level; a precise formulation (e.g., cosine similarity on what embedding?) would aid reproducibility.
  3. [§4.1] Implementation details such as the number of active-learning rounds, batch size per round, and the exact detectors (e.g., Faster R-CNN, YOLO variants) used in each experiment should be consolidated in a single table or paragraph.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. We address each major comment below and outline revisions to improve clarity and rigor.

read point-by-point responses

  1. Referee: [§3] §3 (PAL framework description): the claim that class-specific logistic classifiers trained solely on inference outputs reliably produce useful entropy uncertainty scores rests on an untested assumption about the separability of true/false positives; no ablation is shown on classifier architecture, feature representation (e.g., box coordinates vs. scores), or training-set size, which directly affects the portability and robustness assertions.

    Authors: We acknowledge that the separability assumption underlying the class-specific logistic classifiers merits explicit validation to support the portability claims. The design choice of lightweight logistic regression was motivated by its minimal requirements (only inference outputs) and computational efficiency for black-box detectors, but we agree additional evidence is needed. In the revised manuscript, we will add ablations in Section 3 comparing logistic regression to small neural networks, different feature inputs (scores alone vs. scores plus normalized box coordinates), and varying sizes of the labeled pool used for training the classifiers. These will quantify the robustness of the entropy scores and directly address the concerns about untested assumptions. revision: yes

  2. Referee: [§4] §4 (Experiments): while the abstract and introduction assert 'consistent improvements' across COCO, PASCAL VOC, and BDD100K, the reported results lack explicit quantitative tables or figures showing mAP deltas, label-efficiency curves, and statistical significance against named baselines (e.g., uncertainty sampling, core-set, or diversity-only methods); without these numbers the central empirical claim cannot be evaluated for effect size or reproducibility.

    Authors: We appreciate the referee highlighting the need for more transparent quantitative reporting. The manuscript presents label-efficiency curves and accuracy comparisons across the three datasets, but we agree that explicit mAP delta tables, direct comparisons to named baselines with effect sizes, and statistical significance measures were not sufficiently detailed. In the revision, we will add a summary table in Section 4 reporting mAP at multiple labeling budgets (e.g., 10%, 20%, 50%) with deltas relative to uncertainty sampling, core-set, and diversity-only baselines, include error bars or p-values where feasible, and ensure all curves are clearly labeled. This will allow readers to assess effect sizes and reproducibility more readily. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected in derivation chain

full rationale

The paper describes PAL as a practical, detector-agnostic heuristic that trains lightweight logistic classifiers on inference outputs matched to ground-truth labels from the labeled pool to generate entropy uncertainty scores, then refines selections with image-level entropy, class diversity, and similarity metrics. No equations, self-definitional constructions, fitted parameters renamed as predictions, or load-bearing self-citations appear in the method. The approach relies on conventional active-learning building blocks with external validation via experiments on COCO, PASCAL VOC, and BDD100K, keeping the central claim self-contained and independent of its own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

The framework rests on standard assumptions from active learning and supervised classification without introducing fitted parameters or new entities beyond the named method itself.

axioms (2)
  • domain assumption Logistic regression trained on detector outputs can distinguish true positives from false positives to yield useful uncertainty estimates
    Invoked to generate entropy-based scores for proposals
  • ad hoc to paper Image-level signals (entropy, class diversity, similarity) complement instance uncertainty for better batch selection
    Core selection refinement step described in the abstract
invented entities (1)
  • Portable Active Learning (PAL) framework no independent evidence
    purpose: Detector-agnostic active learning using only inference outputs
    Newly introduced method name and architecture

pith-pipeline@v0.9.0 · 5515 in / 1451 out tokens · 43071 ms · 2026-05-12T04:47:32.390217+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

34 extracted references · 34 canonical work pages · 2 internal anchors

  1. [1]

    Contextual diversity for active learning

    Sharat Agarwal, Himanshu Arora, Saket Anand, and Chetan Arora. Contextual diversity for active learning. InComputer Vision – ECCV 2020: 16th European Conference, Glas- gow, UK, August 23–28, 2020, Proceedings, Part XVI, page 137–153, Berlin, Heidelberg, 2020. Springer-Verlag. 2, 5, 6

    work page 2020

  2. [2]

    Beluch, Tim Genewein, A

    William H. Beluch, Tim Genewein, A. N ¨urnberger, and Jan M. K ¨ohler. The power of ensembles for active learn- ing in image classification.2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 9368– 9377, 2018. 1

    work page 2018

  3. [3]

    Class-balanced active learn- ing for image classification.2022 IEEE/CVF Winter Con- ference on Applications of Computer Vision (WACV), pages 3707–3716, 2021

    Javad Zolfaghari Bengar, Joost van de Weijer, Laura Lopez Fuentes, and Bogdan Raducanu. Class-balanced active learn- ing for image classification.2022 IEEE/CVF Winter Con- ference on Applications of Computer Vision (WACV), pages 3707–3716, 2021. 1, 2

    work page 2022

  4. [4]

    Active learning for deep object detection.arXiv preprint arXiv:1809.09875, 2018

    Clemens-Alexander Brust, Christoph K ¨ading, and Joachim Denzler. Active learning for deep object detection.arXiv preprint arXiv:1809.09875, 2018. doi:10.48550/arXiv.1809.09875. 2

    work page doi:10.48550/arxiv.1809.09875 2018

  5. [5]

    MMDetection: Open MMLab Detection Toolbox and Benchmark

    Kai Chen, Jiaqi Wang, Jiangmiao Pang, Yuhang Cao, Yu Xiong, Xiaoxiao Li, Shuyang Sun, Wansen Feng, Ziwei Liu, Jiarui Xu, Zheng Zhang, Dazhi Cheng, Chenchen Zhu, Tian- heng Cheng, Qijie Zhao, Buyu Li, Xin Lu, Rui Zhu, Yue Wu, Jifeng Dai, Jingdong Wang, Jianping Shi, Wanli Ouyang, Chen Change Loy, and Dahua Lin. MMDetection: Open mmlab detection toolbox and...

    work page Pith review arXiv 1906

  6. [6]

    XGBoost: A scalable tree boosting system

    Tianqi Chen and Carlos Guestrin. XGBoost: A scalable tree boosting system. InProceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 785–794, New York, NY , USA, 2016. ACM. 8

    work page 2016

  7. [7]

    Jiwoong Choi, Ismail Elezi, Hyuk-Jae Lee, Clement Fara- bet, and Jose M. Alvarez. Active learning for deep object detection via probabilistic modeling. In2021 IEEE/CVF In- ternational Conference on Computer Vision (ICCV), pages 10244–10253, 2021. 1

    work page 2021

  8. [8]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Syl- vain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition at scale. InInternational Conference on Learning Representa- tions, 2021. 5

    work page 2021

  9. [9]

    Williams, John Winn, and Andrew Zisserman

    Mark Everingham, Luc Gool, Christopher K. Williams, John Winn, and Andrew Zisserman. The pascal visual object classes (voc) challenge.Int. J. Comput. Vision, 88(2): 303–338, 2010. 6

    work page 2010

  10. [10]

    YOLOX: Exceeding YOLO Series in 2021

    Zheng Ge, Songtao Liu, Feng Wang, Zeming Li, and Jian Sun. Yolox: Exceeding yolo series in 2021.arXiv preprint arXiv:2107.08430, 2021. 6

    work page internal anchor Pith review arXiv 2021

  11. [11]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. InProceed- ings of the IEEE conference on computer vision and pattern recognition, pages 770–778, 2016. 6

    work page 2016

  12. [12]

    Ultralytics yolo11, 2024

    Glenn Jocher and Jing Qiu. Ultralytics yolo11, 2024. 6

    work page 2024

  13. [13]

    Adaptive active learning for image classification

    Xin Li and Yuhong Guo. Adaptive active learning for image classification. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2013. 1

    work page 2013

  14. [14]

    Lawrence Zitnick

    Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Doll ’ar, and C. Lawrence Zitnick. Microsoft coco: Common objects in context. InEuropean Conference on Computer Vision, pages 740–755. Springer International Publishing,

    work page

  15. [15]

    Focal loss for dense object detection.IEEE Transactions on Pattern Analysis and Machine Intelligence, 42(2):318–327, 2020

    Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Doll ´ar. Focal loss for dense object detection.IEEE Transactions on Pattern Analysis and Machine Intelligence, 42(2):318–327, 2020. 6

    work page 2020

  16. [16]

    Nguyen and Arnold Smeulders

    Hieu T. Nguyen and Arnold Smeulders. Active learning us- ing pre-clustering. InProceedings of the Twenty-First In- ternational Conference on Machine Learning, page 79, New York, NY , USA, 2004. Association for Computing Machin- ery. 5

    work page 2004

  17. [17]

    Dinov2: Learning robust visual features with- out supervision, 2024

    Maxime Oquab, Timoth ´ee Darcet, Th ´eo Moutakanni, Huy V o, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Mah- moud Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael Rabbat, Vasu Sharma, Gabriel Synnaeve, Hu Xu, Herv ´e Je- gou, Julien Mairal, ...

    work page 2024

  18. [18]

    Learning transferable visual models from natural language supervision, 2021

    Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, and Ilya Sutskever. Learning transferable visual models from natural language supervision, 2021. 8

    work page 2021

  19. [19]

    Faster r-cnn: towards real-time object detection with region proposal networks

    Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: towards real-time object detection with region proposal networks. InProceedings of the 29th International Conference on Neural Information Processing Systems - Vol- ume 1, page 91–99, Cambridge, MA, USA, 2015. MIT Press. 6

    work page 2015

  20. [20]

    Namboodiri

    Soumya Roy, Asim Unmesh, and Vinay P. Namboodiri. Deep active learning for object detection. InBritish Machine Vision Conference (BMVC), 2018. 2 9

    work page 2018

  21. [21]

    Active learning for con- volutional neural networks: A core-set approach

    Ozan Sener and Silvio Savarese. Active learning for con- volutional neural networks: A core-set approach. InInter- national Conference on Learning Representations (ICLR),

    work page

  22. [22]

    arXiv:1708.00489, doi:10.48550/arXiv.1708.00489. 2, 5, 6

    work page Pith review doi:10.48550/arxiv.1708.00489

  23. [23]

    Very Deep Convolutional Networks for Large-Scale Image Recognition

    Karen Simonyan and Andrew Zisserman. Very deep convo- lutional networks for large-scale image recognition.CoRR, abs/1409.1556, 2014. 6

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  24. [24]

    Cspnet: A new backbone that can enhance learning capability of cnn

    Chien-Yao Wang, Hong-Yuan Mark Liao, I-Hau Yeh, Yueh- Hua Wu, Ping-Yang Chen, and Jun-Wei Hsieh. Cspnet: A new backbone that can enhance learning capability of cnn. 2020 IEEE/CVF Conference on Computer Vision and Pat- tern Recognition Workshops (CVPRW), pages 1571–1580,

    work page 2020

  25. [25]

    Soft teacher for semi-supervised object detection

    Xuancheng Wang, Yifan Zhang, Zhaoyang Zeng, Yuhui Yuan, Jingdong Wang, and Chunhua Shen. Soft teacher for semi-supervised object detection. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 16261–16270, 2021. 3

    work page 2021

  26. [26]

    Visual transform- ers: Token-based image representation and processing for computer vision, 2020

    Bichen Wu, Chenfeng Xu, Xiaoliang Dai, Alvin Wan, Peizhao Zhang, Zhicheng Yan, Masayoshi Tomizuka, Joseph Gonzalez, Kurt Keutzer, and Peter Vajda. Visual transform- ers: Token-based image representation and processing for computer vision, 2020. 5, 8

    work page 2020

  27. [27]

    Entropy-based ac- tive learning for object detection with progressive diversity constraint

    Jiaxi Wu, Jiaxin Chen, and Di Huang. Entropy-based ac- tive learning for object detection with progressive diversity constraint. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022. doi:10.48550/arXiv.2204.07965. 2, 6, 7

    work page doi:10.48550/arxiv.2204.07965 2022

  28. [28]

    Chenhongyi Yang, Lichao Huang, and Elliot J. Crowley. Plug and play active learning for object detection. InPro- ceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2024. arXiv:2211.11612, doi:10.48550/arXiv.2211.11612. 1, 2, 5, 6, 7

    work page doi:10.48550/arxiv.2211.11612 2024

  29. [29]

    Learning loss for active learning

    Donggeun Yoo and In So Kweon. Learning loss for active learning. In2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 93–102, 2019. 1, 2, 7

    work page 2019

  30. [30]

    Bdd100k: A diverse driving dataset for heterogeneous multitask learning

    Fisher Yu, Haofeng Chen, Xin Wang, Wenqi Xian, Yingying Chen, Fangchen Liu, Vashisht Madhavan, and Trevor Dar- rell. Bdd100k: A diverse driving dataset for heterogeneous multitask learning. In2020 IEEE/CVF Conference on Com- puter Vision and Pattern Recognition (CVPR), pages 2633– 2642, 2018. 6

    work page 2018

  31. [31]

    Multiple in- stance active learning for object detection

    Tianning Yuan, Fang Wan, Mengying Fu, Jianzhuang Liu, Songcen Xu, Xiangyang Ji, and Qixiang Ye. Multiple in- stance active learning for object detection. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021. 1, 2, 6, 7

    work page 2021

  32. [32]

    State-relabeling adversarial active learning

    Beichen Zhang, Liang Li, Shijie Yang, Shuhui Wang, Zheng- Jun Zha, and Qingming Huang. State-relabeling adversarial active learning. In2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 8753–8762,

    work page

  33. [33]

    Active teacher for semi-supervised object detection

    Yifan Zhang, Xuancheng Wang, Zhaoyang Zeng, Yuhui Yuan, Jingdong Wang, and Chunhua Shen. Active teacher for semi-supervised object detection. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022. 3

    work page 2022

  34. [34]

    Generative Adversarial Active Learning

    Jia-Jie Zhu and Jos ´e Bento. Generative adversarial ac- tive learning.arXiv preprint arXiv:1702.07956, 2017. doi:10.48550/arXiv.1702.07956. 2, 5 10

    work page Pith review doi:10.48550/arxiv.1702.07956 2017