Ten Years of Active Learning Techniques and Object Detection: A Systematic Review
Open accessThe DOI matches its Crossref record; 112 of the 113 references it cites were found in a registry or an index; 42 of the 113 sources its claims cite were checked for support.
Abstract
Object detection (OD) coupled with active learning (AL) has emerged as a powerful synergy in the field of computer vision, harnessing the capabilities of machine learning (ML) to automatically identify and perform image-based objects localisation while actively engaging human expertise to iteratively enhance model performance and foster machine-based knowledge expansion. Their prior success, demonstrated in a wide range of fields (e.g., industry and medicine), motivated this work, in which a comprehensive and systematic review of OD and AL techniques was carried out, considering reputed technical/scientific publication databases—such as ScienceDirect, IEEE, PubMed, and arXiv—and a temporal range between 2010 and December 2022. The primary inclusion criterion for papers in this review was the application of AL techniques for OD tasks, regardless of the field of application. A total of 852 articles were analysed, and 60 articles were included after full screening. Among the remaining ones, relevant topics such as AL sampling strategies used for OD tasks and groups categorisation can be found, along with details regarding the deep neural network architectures employed, application domains, and approaches used to blend learning techniques with those sampling strategies. Furthermore, an analysis of the geographical distribution of OD researchers across the globe and their affiliated organisations was conducted, providing a comprehensive overview of the research landscape in this field. Finally, promising research opportunities to enhance the AL process were identified, including the development of novel sampling strategies and their integration with different learning techniques.
Overview
CoChat Summary
## TL;DR
This systematic review surveys ten years of active learning (AL) methods for object detection (OD), covering 852 records and including 60 full-text papers published between 2010 and December 2022. It finds that uncertainty-based methods are most common, with ensembles and Monte-Carlo dropout reported as the strongest performers overall.
## Key findings
- The review screened 852 records, removed 4 duplicates, and retained 60 full-text documents for the final set of papers.
- The literature is concentrated in 2D OD and in countries such as China, the United States, and Germany; public datasets like PASCAL, MS-COCO, and KITTI were extensively used.
- Uncertainty-based query strategies are the most explored in 2D OD, followed by hybrid methods; the paper also notes that uncertainty-based methods are prone to data bias because selected samples are “hardly representative” of the full unlabeled dataset.
- The paper states that ensemble- and Monte-Carlo-dropout-based methods “have shown the best results, overall,” and that balancing uncertainty and diversity can yield more robust methods.
## Methods
- Systematic review following the PRISMA procedure, focused on AL methods combined with image-based OD.
- Search covered Science Direct, IEEE Xplore, arXiv, and PubMed; keywords combined “active learning” and “object detection” with acquisition/query/uncertainty/diversity/sampling terms.
- Inclusion criteria limited studies to 2010–December 2022, peer-reviewed journal or conference papers, English language, and AL applied to OD in 2D or 3D; extracted data included publication metadata, AL scenario, query strategy, OD architecture, application field, metrics, and AL category.
## Limitations
- The paper itself notes that the review is focused on OD and does not cover segmentation or classification applications.
- It also states that many areas remain under-explored, including healthcare, document layout detection, pose estimation, and surface defect detection; this is an evident coverage limitation of the underlying literature, not necessarily of the review design.
- Evident from the text: the synthesis is largely descriptive and taxonomy-based, so it does not report a quantitative meta-analysis of effect sizes or pooled performance estimates.
Model-written summary. Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
21 Valid · 14 Not valid · 31 Not checked · 52 Own results and interpretations
Some cited sources support this claim and some don’t — read each source’s verdict below.
In the paper
“Other applications such as document layout detection [14,17], pose estimation [7], and surface defect detection [55] remain under-explored, as well.”
What the sources say
- Partly supportedcites [14] Shen, Z., Zhao, J., Dell, M., Yu, Y., and Li, W. (2020). OLALA: Object-Level Active Learning Based Layout Annotation. arXiv. sourceExperimental — accuracy not yet measured
The Abstract and Introduction identify document layout detection as the application of OLALA and explain why existing image-level active-learning methods are poorly suited to densely arranged document regions. They do not establish that document layout detection generally remains under-explored. This citation is attached to document layout detection, not to pose estimation or surface defect detection. (LLM, conf=0.78)
- Supportedcites [17] Boillet, M., Kermorvant, C., and Paquet, T. (2022). Confidence Estimation for Object Detection in Document Images. arXiv. source
The citation is attached to document layout detection as an example of an application. The Experimental setup (Section 3; page unavailable) tests confidence estimators on physical document page detection and handwritten text line detection, supporting that identification. (LLM, conf=0.87)
- Disputedcites [7] Lee, J., Balachandran, R., Kondak, K., Coelho, A., De Stefano, M., Humt, M., Feng, J., Asfour, T., and Triebel, R. (2022). Virtual Reality via Object Pose Estimation and Active Learning: Realizing Telepresence Robots with Aerial Manipulation Capabilities. arXiv. sourceExperimental — accuracy not yet measured
For the pose-estimation portion cited to this paper, Related Work (p. 5) says the literature is vast and points to a comprehensive survey. The paper studies pose estimation, but it does not characterize the field as under-explored; it says the novelty lies in solutions tailored to its VR application. (LLM, conf=0.85)
Checked by a reasoning model — may be wrong.
- Supportedcites [55] Lv, X., Duan, F., Jiang, J.J., Fu, X., and Gan, L. (2020). Deep Active Learning for Surface Defect Detection. Sensors, 20. source
The Introduction (page unavailable) says existing active-learning strategies cannot be directly applied to defect detection and that active learning for defect detection still faces major challenges. The paper also develops an active-learning framework for metallic surface defect detection, supporting the cited example as an under-explored application of active learning. (LLM, conf=0.84)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Furthermore, previous reviews [8,10-13] are more oriented to other AI areas such as natural language processing, speech processing, and computer vision, but without going deeper into OD, showing only a few related papers.”
What the sources say
- Partly supportedcites [8] Ren, 2020, A Survey of Deep Active Learning, ACM Comput. Surv. sourceExperimental — accuracy not yet measured
Our copy confirms the first half of what is attributed: Ren et al.'s survey is organised around applications in vision, NLP and speech/audio (Section 4 heading, p. 21: 'visual data processing (such as object detection, semantic segmentation, etc.), NLP ..., speech and audio processing'), and Section 4.3 (p. 26) states DeepAL applications are 'mainly focused on visual image processing tasks, although there are also applications in NLP and other fields'. What the copy does not support is the comparative judgement that it does not 'go deeper into OD, showing only a few related papers': the paper never assesses prior reviews that way (it instead claims to be 'the first comprehensive review work in the field of deep active learning', Section 1, p. 1 and Section 6, p. 28), and object detection is given an explicit subsection (4.1.2, pp. 22-23) plus Table 2 rows, discussing several named DeepAL detection frameworks ([4], [63], [170], [178], [140]). So the breadth part holds while the 'only a few papers / no depth' characterisation is the citing authors' own gloss, not a statement of this source. (LLM, conf=0.65)
- Partly supportedcites [10] Wu, 2022, A survey of human-in-the-loop for machine learning, Future Gener. Comput. Syst. sourceExperimental — accuracy not yet measured
The Introduction (§1.3) says this survey discusses natural language processing and computer vision, and the computer-vision summary (Table 3, p. 10) lists only two object-detection studies. That supports the claim that object detection is a small part of a broader review. Our copy does not show speech processing as a comparable focus, and the survey does include a dedicated object-detection subsection (§3.2.1). (LLM, conf=0.83)
- Partly supportedcites [11] Budd, S., Robinson, E.C., and Kainz, B. (2019). A Survey on Active Learning and Human-in-the-Loop Deep Learning for Medical Image Analysis. arXiv. sourceExperimental — accuracy not yet measured
The introduction (p. 2) identifies this as a review of human-in-the-loop deep learning for medical image analysis and says it focuses on medical image segmentation, supporting the point that it is not an in-depth review of object detection (OD). Our copy does not establish that it shows only a few OD-related papers. The references to natural language and speech processing appear to describe the cited reviews collectively, not necessarily this source. (LLM, conf=0.78)
- Partly supportedcites [12] Abdar, 2021, A review of uncertainty quantification in deep learning: Techniques, applications and challenges, Inf. Fusion sourceExperimental — accuracy not yet measured
The review covers natural language processing (Section 8.3, p. 28) and computer vision (Section 8.1, p. 26), while its computer-vision overview cites only three object-detection studies. This supports the claim that object detection is not its main focus, but the passages do not establish the claimed orientation toward speech processing. (LLM, conf=0.82)
- Partly supportedcites [13] Takezoe, R., Liu, X., Mao, S., Chen, M.T., Feng, Z., Zhang, S., and Wang, X. (2022). Deep Active Learning for Computer Vision: Past and Future. arXiv. sourceExperimental — accuracy not yet measured
Our copy supports the breadth half of the attribution: Takezoe et al. (2023) is a DeepAL survey oriented to computer vision as a whole, with the Introduction (p. 2) noting that existing surveys cover 'broad AI research including vision, nature language processing and machine learning' and the Conclusion (p. 13) listing image classification, object detection, segmentation and video recognition as its covered tasks — so it is not an OD-focused review. It does not support the rest: the paper reviews no speech-processing literature and mentions text classification/sentiment analysis only in passing (Introduction, p. 1), and it does go into object detection in a dedicated Section III.B (pp. 6-7) with two subsections ('Query Strategy', 'Annotation Reduction') covering roughly fifteen OD-specific works (refs [76]-[90]), which is more than 'only a few related papers'. (LLM, conf=0.72)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
Every source the paper cites for this claim was checked and supports it.
In the paper
“The most relevant impacts of AL methods were observed from 2014, after the releasing of architectures such as YOLO [23], Faster R-CNN [33], among others, which attracted the interest of researchers and practitioners in applying AL in OD, as can be concluded from Figure 3.”
What the sources say
- Supportedcites [23] Redmon, J., Divvala, S.K., Girshick, R.B., and Farhadi, A. (2015). You Only Look Once: Unified, Real-Time Object Detection. arXiv. source
The Abstract (p. 1) presents YOLO as a new approach to object detection, and the Conclusion (p. 8) identifies it as a unified object-detection model. Citation [23] is attached specifically to “YOLO”; the sentence’s claim about active-learning trends is attributed to the citing paper’s Figure 3, not to this source. (LLM, conf=0.93)
- Supportedcites [33] Ren, 2015, Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks, IEEE Trans. Pattern Anal. Mach. Intell. source
The citation identifies Faster R-CNN as an object-detection architecture, and the Related Work section (p. 3, Figure 2) describes it as a unified network for object detection. The sentence attributes the conclusion about active-learning trends to the citing paper’s Figure 3, not to this source. (LLM, conf=0.94)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Supported
The sources with a verdict support this claim; the rest weren’t checked yet, have no full text we could read, or their verdicts aren’t rated yet.
In the paper
“In the health area, four articles articles were found [37,57-59]”
What the sources say
- Not checked — no full text of the sourcecites [37] Kee, 2018, Query-by-committee improvement with diversity and density in batch active learning, Inf. Sci. source
- Not checked — no full text of the sourcecites [57] Tang, 2021, QBox: Partial Transfer Learning With Active Querying for Object Detection, IEEE Trans. Neural Netw. Learn. Syst. source
- Not checked — no full text of the sourcecites [58] Gu, H., Haeri, M., Ni, S., Williams, C.K., Zarrin-Khameh, N., Magaki, S., and Chen, X.A. (2022). Detecting Mitoses with a Convolutional Neural Network for MIDOG 2022 Challenge. arXiv. source
- Supportedcites [59] Yun, J.B., Oh, J., and Yun, I.D. (2020). Gradually Applying Weakly Supervised and Active Learning for Mass Detection in Breast Ultrasound Images. arXiv. source
This citation identifies one of the four articles the citing authors classify as being in the health area; it does not attribute the count of four to this paper. The Introduction (p. 1) concerns breast-cancer detection using ultrasound, and the Datasets section (p. 4) describes breast-ultrasound images from hospital patients. (LLM, conf=0.90)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“The use of public databases such as PASCAL [72], MS-COCO [73], or autonomous driving [74-76] simplifies the performance of experiments to compare sampling methods.”
What the sources say
- Partly supportedcites [72] Everingham, 2015, The Pascal Visual Object Classes Challenge: A Retrospective, Int. J. Comput. Vis. sourceExperimental — accuracy not yet measured
The discussion of standard assessment software (section 6.2.1; page not available) says that VOC’s data splits, evaluation protocol, and code enabled results to be compared consistently. This supports the point for PASCAL, the example linked to this citation, but the paper does not specifically say that the database simplifies experiments comparing sampling methods. (LLM, conf=0.84)
- Supportedcites [73] Lin, T.Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P., and Zitnick, C.L. (2014, January 6–12). Microsoft COCO: Common objects in context. Proceedings of the Computer Vision—ECCV 2014, Zurich, Switzerland. source
The citation is attached to MS-COCO as an example of a public database, and this paper introduces that dataset. The Algorithmic Analysis (p. 11) describes its training, validation, and test splits and says the authors will release the training and validation annotations. The broader point about comparing sampling methods is not specifically attributed to this citation. (LLM, conf=0.90)
- Not checked — no full text of the sourcecites [74] Geiger, A., Lenz, P., and Urtasun, R. (2012, January 16–21). Are we ready for autonomous driving? The KITTI vision benchmark suite. Proceedings of the 2012 IEEE Conference on Computer Vision and Pattern Recognition, Providence, RI, USA. source
- Not checked — no full text of the sourcecites [75] Sun, P., Kretzschmar, H., Dotiwalla, X., Chouard, A., Patnaik, V., Tsui, P., Guo, J., Zhou, Y., Chai, Y., and Caine, B. (2019). Scalability in Perception for Autonomous Driving: Waymo Open Dataset. arXiv. source
- Supportedcites [76] Caesar, H., Bankiti, V., Lang, A.H., Vora, S., Liong, V.E., Xu, Q., Krishnan, A., Pan, Y., Baldan, G., and Beijbom, O. (2019). nuScenes: A multimodal dataset for autonomous driving. arXiv. source
The Introduction identifies nuScenes as a benchmark dataset for autonomous driving. Section 1.1 (Contributions) says the data and evaluation code are available online for research, supporting its use as a public-dataset example. The citation need not establish the citing authors’ separate point about comparing sampling methods. (LLM, conf=0.90)
plainstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Partly supported
27 more claims · 5 with their cited sources checked
Create a free account to see all 118 claims and to verify the 26 references still open.
- Probably not a substantial claim. 68
- It may state two separate claims as one. 30
Citations
Cites
ImageNet classification with deep convolutional neural networks
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2017) · Communications of the ACM · doi:10.1145/3065386 · Cited by 112,155
Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
Ren, S., He, K., Girshick, R., & Sun, J. (2017) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2016.2577031 · Cited by 73,472
You Only Look Once: Unified, Real-Time Object Detection
Redmon, J., Divvala, S., Girshick, R., & Farhadi, A. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.91 · Cited by 53,803
Microsoft COCO: Common Objects in Context
Lin, T.-Y., Maire, M., Belongie, S., Hays, J., Perona, P., Ramanan, D., Dollár, P., & Zitnick, C. L. (2014) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-10602-1_48 · Cited by 45,460
Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation
Girshick, R., Donahue, J., Darrell, T., & Malik, J. (2014) · 2014 IEEE Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2014.81 · Cited by 32,684
Focal Loss for Dense Object Detection
Lin, T.-Y., Goyal, P., Girshick, R., He, K., & Dollar, P. (2017) · 2017 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2017.324 · Cited by 30,581
Girshick, R. (2015) · 2015 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2015.169 · Cited by 28,105
SSD: Single Shot MultiBox Detector
Liu, W., Anguelov, D., Erhan, D., Szegedy, C., Reed, S., Fu, C.-Y., & Berg, A. C. (2016) · Lecture Notes in Computer Science · doi:10.1007/978-3-319-46448-0_2 · Cited by 22,128
YOLO9000: Better, Faster, Stronger
Redmon, J., & Farhadi, A. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.690 · Cited by 19,892
YOLOv3: An Incremental Improvement
Redmon, J., & Farhadi, A. (2018) · arXiv · doi:10.48550/arxiv.1804.02767 · Cited by 17,141
Research works printed but not linked: 2
- Shen, Z., Zhao, J., Dell, M., Yu, Y., and Li, W. (2020). OLALA: Object-Level Active Learning Based Layout Annotation. arXiv.
- Liu, W., Anguelov, D., Erhan, D., Szegedy, C., Reed, S., Fu, C.Y., and Berg, A.C. (2016). Computer Vision—ECCV 2016, Springer International Publishing.
Cited by
Ribeiro, M., Santos, J., Lobo, J., Araújo, S., Magalhães, L., & Adão, T. (2024) · international conference on 3D web technology · doi:10.1145/3665318.3677172 · Cited by 57
“This iterative process – known as active learning with the human in the loop [Garcia et al., 2023] – enhances the system’s accuracy and efficiency, ultimately allowing the computational components to achieve a high level of reliability and autonomy in managing museum data.”
Self-citationActive learning for industrial defect detection: a study on hybrid sampling strategies
Gonzalez, D. G., Nascimento, R. C., Rocha, C. D., Silva, M. F., Filipe, V. M. D. J., Rocha, L. F., Magalhães, L., & Cunha, A. M. T. D. S. (2026) · The International Journal of Advanced Manufacturing Technology · doi:10.1007/s00170-025-17378-7 · Cited by 2
Self-citationRarity-Aware Stratified Active Learning for Class-Imbalanced Industrial Object Detection
Benhafid, Z., & Selouani, S. (2026) · Applied Sciences · doi:10.3390/app16031236 · Cited by 3
“Based on the adopted query strategy, AL approaches are commonly categorized into three main families [24,25]: uncertainty-based, diversity-based, and hybrid uncertainty–diversity methods.”
Improving cross-site generalization in construction object detection via hard negative mining
Seong, J., Kim, H.-S., & Jung, H.-J. (2026) · Automation in Construction · doi:10.1016/j.autcon.2026.106761 · Cited by 2
Yang, H., Wang, Y., & Xin, H. (2025) · Measurement · doi:10.1016/j.measurement.2025.118933 · Cited by 2
Metadata
- Authors
- Garcia, D., Carias, J., Adão, T., Jesus, R., Cunha, A., & Magalhães, L. G.
- Affiliations
- Centro de Computação Gráfica—CCG/zgdv, University of Minho, Campus de Azurém, Edifício 14, 4800-058 Guimarães, Portugal
- Centro de Computação Gráfica (PT)
- University of Trás-os-Montes and Alto Douro (PT)
- Universidade da Coruña (ES)
- Department of Engineering, School of Sciences and Technology, University of Trás-os-Montes e Alto Douro, 5000-801 Vila Real, Portugal
- ALGORITMI Research Centre/LASI, University of Minho, 4800-058 Guimarães, Portugal
- Venue
- Applied Sciences
- Publisher
- MDPI AG
- Published
- 2023-09-25
- Volume
- 13
- Issue
- 19
- Pages
- 10667
- ISSN
- 2076-3417 (ISSN-L)
- OpenAlex
- W4387047685
- Semantic Scholar
- 06db8e69ceb7cc0d6d47ff7ce92bc56826ae8248
- Cited by
- 17
- References
- 74
- Citations by year
- 2026: 10 · 2025: 5 · 2024: 1 · 2023: 1
- FWCI
- 0.44
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- Applied Sciences (publishedVersion, cc-by)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Portuguese National Funding Agency for Science, Research and Technology (RCAAP Project by FCT) (submittedVersion, cc-by-nc-nd)
- Funders
- Mobilizing Agendas for Business Innovation (C644937233-00000047, C644866286-00000011)
- Mobilizing Agendas for Reindustrialization (C644937233-00000047, C644866286-00000011)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning and Algorithms
- Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
- Keywords
- active learning (machine learning), object detection, sampling strategy, deep neural network, human-in-the-loop learning, research trends, research opportunities
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol and qwen3.8-omni-flash).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Ten Years of Active Learning Techniques and Object Detection: A Systematic Review. https://app.cochat.ai/papers/10.3390/app131910667
- The paper
- Garcia, D., Carias, J., Adão, T., Jesus, R., Cunha, A., & Magalhães, L. G. (2023). Ten Years of Active Learning Techniques and Object Detection: A Systematic Review. Applied Sciences. https://doi.org/10.3390/app131910667
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.