The DOI matches its Crossref record; 47 of the 49 references it cites were found in a registry or an index.
Active Learning for Deep Object Detection via Probabilistic Modeling
Open accessAbstract
Active learning aims to reduce labeling costs by selecting only the most informative samples on a dataset. Few existing works have addressed active learning for object detection. Most of these methods are based on multiple models or are straightforward extensions of classification methods, hence estimate an image’s informativeness using only the classification head. In this paper, we propose a novel deep active learning approach for object detection. Our approach relies on mixture density networks that estimate a probabilistic distribution for each localization and classification head’s output. We explicitly estimate the aleatoric and epistemic uncertainty in a single forward pass of a single model. Our method uses a scoring function that aggregates these two types of uncertainties for both heads to obtain every image’s informativeness score. We demonstrate the efficacy of our approach in PASCAL VOC and MS-COCO datasets. Our approach outperforms single-model based methods and performs on par with multi-model based methods at a fraction of the computing cost. Code is available at https://github.com/NVlabs/AL-MDN.
Overview
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source doesn’t support the claim as written.
In the citing paper
“In the last one, where we use 7 , 000 samples, 5 , 000 of which are actively labeled, our method outperforms the random baseline by 9 . 1% and the best existing active learning methods by more than 2 . 8% [5].”
From Not All Labels Are Equal: Rationalizing The Labeling Costs for Training Object Detection (2021) · cites this paper as [5]
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“For MS-COCO, in Fig. 4a-b, we observe that in the first active learning cycle, our method outperforms the random baseline by 5 . 8% , the best-performing AL method by 2 . 7% [5], the semi-supervised method by 5 . 4% , and the ensembles by 5% .”
From Not All Labels Are Equal: Rationalizing The Labeling Costs for Training Object Detection (2021) · cites this paper as [5]
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“We observe that this difference is maintained in the other cycles, including the last active learning cycle where our method outperforms the semisupervised method, multi-model methods [9, 12] and the best AL method [5] by 1 . 6% .”
From Not All Labels Are Equal: Rationalizing The Labeling Costs for Training Object Detection (2021) · cites this paper as [5]
The cited source says what the claim attributes to it.
In the citing paper
“In the second cycle, our approach outperforms all the other methods, including PM [5], by almost 4% or more.”
From Not All Labels Are Equal: Rationalizing The Labeling Costs for Training Object Detection (2021) · cites this paper as [5]
The cited source says what the claim attributes to it.
In the citing paper
“Compared to state-of-the-art methods [4-8], our approach yields up to a 7 . 7% and 7% relative mAP improvement for PASCAL-VOC and MS-COCO, respectively.”
From Not All Labels Are Equal: Rationalizing The Labeling Costs for Training Object Detection (2021) · cites this paper as [5]
The cited source says what the claim attributes to it.
In the citing paper
“The random sampling strategy is mostly used for comparison purposes, grounded on to the principle that AL-based methods have the potential to perform better (e.g., [5,48,52,60]).”
From Ten Years of Active Learning Techniques and Object Detection: A Systematic Review (2023) · cites this paper as [48]
The cited source says what the claim attributes to it.
In the citing paper
“The classifiers used for uncertainty estimation ignore the location information [48].”
From Ten Years of Active Learning Techniques and Object Detection: A Systematic Review (2023) · cites this paper as [48]
The cited source says what the claim attributes to it.
In the citing paper
“White methods, based on multi-models (like classifier assemblies which increase the number of parameters [48]), have a high computing cost, while black methods may not need many parameters.”
From Ten Years of Active Learning Techniques and Object Detection: A Systematic Review (2023) · cites this paper as [48]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Deep Residual Learning for Image Recognition
He, K., Zhang, X., Ren, S., & Sun, J. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.90 · Cited by 235,474 · cited 1×
“We compare the performance of SSD using different backbones such as Resnet-34 and Resnet-50 [17], and Faster-RCNN [32] detector trained using our actively sampled dataset.”
MethodVery Deep Convolutional Networks for Large-Scale Image Recognition
Simonyan, K., & Zisserman, A. (2014) · arXiv · doi:10.48550/arxiv.1409.1556 · Cited by 114,148 · cited 1×
“Experimental settings: We employ Single Shot MultiBox Detector (SSD) [31], which is widely used in active learning studies [33, 40], with a VGG-16 backbone [37].”
MethodFaster 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
A Mathematical Theory of Communication
Shannon, C. E. (1948) · The MIT Press eBooks · doi:10.1002/j.1538-7305.1948.tb00917.x · Cited by 50,387
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
(2016) · Cited by 31,590
Feature Pyramid Networks for Object Detection
Lin, T.-Y., Dollar, P., Girshick, R., He, K., Hariharan, B., & Belongie, S. (2017) · 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2017.106 · Cited by 30,939 · cited 3×
“21 Faster-RCNN [32] Resnet-50-FPN [29] 72.”
MethodInfluentialGirshick, R. (2015) · 2015 IEEE International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2015.169 · Cited by 28,105
Auto-Encoding Variational Bayes
Kingma, D. P., & Welling, M. (2013) · arXiv · doi:10.48550/arxiv.1312.6114 · Cited by 20,067 · cited 1×
“2, and obtain the class probability distribution for the k-th mixture by using the reparameterization trick [25] of applying Gaussian noise and variance Σp to μ k p [24]:”
MethodThe pascal visual object classes (VOC) challenge
Everingham, M., Van Gool, L., Williams, C. K. I., Winn, J., & Zisserman, A. (2010) · International Journal of Computer Vision · doi:10.1007/s11263-009-0275-4 · Cited by 19,599 · cited 2×
“We demonstrate the benefits of our approach on PASCAL VOC [9] and MS-COCO [30] in a single-stage architecture such as SSD [31], and show generalized performance in a two-stage architecture such as Faster-RCNN [32].”
MethodDropout as a bayesian approximation: Representing model uncertainty in deep learning
Gal, Y., & Ghahramani, Z. (2015) · arXiv · doi:10.48550/arxiv.1506.02142 · Cited by 6,471 · cited 7×
“However, our method uses a single forward pass of a single model to estimate the uncertainties, which is more efficient than ensembles and MC-dropout based methods.”
BackgroundInfluentialSqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size
Iandola, F. N., Han, S., Moskewicz, M. W., Ashraf, K., Dally, W. J., & Keutzer, K. (2016) · arXiv · doi:10.48550/arxiv.1602.07360 · Cited by 6,201 · cited 1×
“In general, a fast forward (backward) step and a small model size leads to a lower training cost and data sampling time during active learning [18, 21].”
BackgroundSimple and Scalable Predictive Uncertainty Estimation using Deep Ensembles
Lakshminarayanan, B., Pritzel, A., & Blundell, C. (2016) · arXiv · doi:10.48550/arxiv.1612.01474 · Cited by 3,666 · cited 1×
“based methods, the number of parameters is proportional to the number of SSD models in the ensemble [26].”
MethodWhat Uncertainties Do We Need in Bayesian Deep Learning for Computer Vision?
Kendall, A., & Gal, Y. (2017) · arXiv · doi:10.48550/arxiv.1703.04977 · Cited by 3,088 · cited 2×
“2, and obtain the class probability distribution for the k-th mixture by using the reparameterization trick [25] of applying Gaussian noise and variance Σp to μ k p [24]:”
BackgroundMethodAleatoric and epistemic uncertainty in machine learning: an introduction to concepts and methods
Hüllermeier, E., & Waegeman, W. (2021) · Machine Learning · doi:10.1007/s10994-021-05946-3 · Cited by 1,834 · cited 1×
“tant in active learning, as it allows the deep learning models to know about their limitations [21, 38], i.”
Background(1994) · Aston Publications Explorer (Aston University) · Cited by 1,245
Settles, B. (2012) · Cited by 1,183
Active Learning for Convolutional Neural Networks: A Core-Set Approach
Sener, O., & Savarese, S. (2017) · arXiv · doi:10.48550/arxiv.1708.00489 · Cited by 922
Deep Bayesian Active Learning with Image Data
Gal, Y., Islam, R., & Ghahramani, Z. (2017) · arXiv · doi:10.48550/arxiv.1703.02910 · Cited by 900 · cited 1×
“To compute these types of uncertainty, researchers use multi-model based approaches, such as ensembles [2] or Monte Carlo (MC) dropout [13].”
MethodBatchBALD: Efficient and Diverse Batch Acquisition for Deep Bayesian Active LearningPreprint
Kirsch, A., van Amersfoort, J. R., & Gal, Y. (2019) · Neural Information Processing Systems · doi:10.48550/arxiv.1906.08158 · Cited by 826
Learning Loss for Active Learning
Yoo, D., & Kweon, I. S. (2019) · 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2019.00018 · Cited by 740 · cited 13×
“Numbers are taken from [40]; b) Comparison to multiple model-based methods, ensemble and MC-dropout.”
BackgroundMethodInfluentialThe Power of Ensembles for Active Learning in Image Classification
Beluch, W. H., Genewein, T., Nurnberger, A., & Kohler, J. M. (2018) · 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2018.00976 · Cited by 655 · cited 6×
“For ensembles, we follow [2], building an ensemble of three independent models.”
BackgroundMethodInfluentialSettles, B. (2012) · Synthesis Lectures on Artificial Intelligence and Machine Learning · doi:10.2200/s00429ed1v01y201207aim018 · Cited by 605
Choi, J., Chun, D., Kim, H., & Lee, H.-J. (2019) · 2019 IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2019.00059 · Cited by 505 · cited 2×
“In experiments, we set η to 2 as suggested in [6].”
BackgroundMethodSelf-citationActive Learning for Convolutional Neural Networks: A Core-Set Approach
Şener, O., & Savarese, S. (2018) · International Conference on Learning Representations · Cited by 460
Kwon, Y., Won, J.-H., Kim, B. J., & Paik, M. C. (2020) · Computational Statistics & Data Analysis · doi:10.1016/j.csda.2019.106816 · Cited by 424 · cited 1×
“In this case, ual is C × C matrix where the value of each diagonal element can be interpreted as a class-specific aleatoric uncertainty [27].”
BackgroundDeep Bayesian Active Learning with Image Data
Ghahramani, Z., Gal, Y., & Islam, R. (2017) · PMLR · doi:10.17863/cam.11070 · Cited by 392
Feng, D., Rosenbaum, L., & Dietmayer, K. (2018) · IEEE International Conference on Intelligent Transportation Systems (ITSC) · doi:10.1109/itsc.2018.8569814 · Cited by 274
Hora, S. C. (1996) · Reliability Engineering & System Safety · doi:10.1016/s0951-8320(96)00077-4 · Cited by 249
Single-Model Uncertainties for Deep LearningPreprint
Tagasovska, N., & López-Paz, D. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1811.00908 · Cited by 211
Active Learning for Deep Detection Neural Networks
Aghdam, H. H., Gonzalez-Garcia, A., Lopez, A., & Weijer, J. (2019) · 2019 IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2019.00377 · Cited by 148
BayesOD: A Bayesian Approach for Uncertainty Estimation in Deep Object Detectors
Harakeh, A., Smart, M. H. W., & Waslander, S. L. (2020) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra40945.2020.9196544 · Cited by 130 · cited 1×
“In general, the predictive uncertainty is decomposed into aleatoric and epistemic uncertainty [15, 20].”
BackgroundLocalization-Aware Active Learning for Object Detection
Kao, C.-C., Lee, T.-Y., Sen, P., & Liu, M.-Y. (2019) · Lecture Notes in Computer Science · doi:10.1007/978-3-030-20876-9_32 · Cited by 118
Choi, S., Lee, K., Lim, S., & Oh, S. (2018) · Proceedings - IEEE International Conference on Robotics and Automation/Proceedings · doi:10.1109/icra.2018.8462978 · Cited by 111
Scalable Active Learning for Object Detection
Haussmann, E., Fenzi, M., Chitta, K., Ivanecky, J., Xu, H., Roy, D., Mittel, A., Koumchatzky, N., Farabet, C., & Alvarez, J. M. (2020) · 2020 IEEE Intelligent Vehicles Symposium (IV) · doi:10.1109/iv47402.2020.9304793 · Cited by 110 · cited 8×
“These methods reach good results but come with several limitations [11, 16].”
BackgroundMethodInfluentialSelf-citationDeep Active Learning for Efficient Training of a LiDAR 3D Object Detector
Feng, D., Wei, X., Rosenbaum, L., Maki, A., & Dietmayer, K. (2019) · 2019 IEEE Intelligent Vehicles Symposium (IV) · doi:10.1109/ivs.2019.8814236 · Cited by 85 · cited 6×
“These methods reach good results but come with several limitations [11, 16].”
BackgroundMethodInfluentialHighlander, T. C., & Rodríguez, A. (2015) · British Machine Vision Conference (BMVC) · doi:10.5244/c.29.160 · Cited by 75 · cited 1×
“In general, a fast forward (backward) step and a small model size leads to a lower training cost and data sampling time during active learning [19, 22].”
BackgroundDeep active learning for object detection
Roy, S. D., Unmesh, A., & Namboodiri, V. P. (2018) · Cited by 71
Mixture Dense Regression for Object Detection and Human Pose Estimation
Varamesh, A., & Tuytelaars, T. (2020) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr42600.2020.01310 · Cited by 50 · cited 3×
“The works of [18, 39] attempt to solve a multimodal regression task.”
BackgroundLarge-Scale Visual Active Learning with Deep Probabilistic EnsemblesPreprint
Chitta, K., Álvarez, J. M., & Lesnikowski, A. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1811.03575 · Cited by 38
Deep Mixture Density Network for Probabilistic Object Detection
He, Y., & Wang, J. (2020) · IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) · doi:10.1109/iros45743.2020.9340882 · Cited by 22
Adaptive Semantic Segmentation with a Strategic Curriculum of Proxy LabelsPreprint
Chitta, K., Feng, J., & Hebert, M. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1811.03542 · Cited by 12
Less is More: An Exploration of Data Redundancy with Active Dataset Subsampling.Preprint
Chitta, K., Álvarez, J. M., Haußmann, E., & Farabet, C. (2019) · arXiv (Cornell University) · Cited by 12
Self-citationTask Agnostic Robust Learning on Corrupt Outputs by Correlation-Guided Mixture Density Networks
Choi, S., Hong, S., Lee, K., & Lim, S. (2020) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr42600.2020.00393 · Cited by 8 · cited 2×
“Additionally, all these studies do not take into account both types of uncertainty coming from the bounding box regression and the classification tasks [7, 8, 18, 39, 41].”
BackgroundDeep Multivariate Mixture of Gaussians for Object Detection under Occlusion
He, Y., & Wang, J. (2019) · Cited by 3
Mixture-Model-based Bounding Box Density Estimation for Object DetectionPreprint
Yoo, J., Seo, G., Chung, I., & Kwak, N. (2019) · arXiv (Cornell University) · Cited by 2
Density-Based Object Detection: Learning Bounding Boxes without Ground Truth AssignmentPreprint
Yoo, J., Lee, H., Chung, I., Seo, G., & Kwak, N. (2021) · arXiv (Cornell University) · Cited by 2
Self-citation
Research works printed but not linked: 4
- 10.5555/3295222.3295309 · DOI
- 10.48550/arXiv.1405.0312 · DOI · Paper page
- 10.1007/978-3-319-46448-0_2 · DOI · Paper page
- 10.1145/584091.584093 · DOI
Cited by
Sorted by citations
Tiny object detection with context enhancement and feature purification
Xiao, J., Guo, H., Zhou, J. Y., Zhao, T., Yu, Q., Chen, Y., & Wang, Z. (2022) · Expert Systems with Applications · doi:10.1016/j.eswa.2022.118665 · Cited by 153
A comprehensive survey on deep active learning in medical image analysis
Wang, H., Jin, Q., Li, S., Liu, S., Wang, M., & Song, Z. (2024) · Medical Image Analysis · doi:10.1016/j.media.2024.103201 · Cited by 157
Wang, K., Shen, C., Li, X., & Lǚ, J. (2025) · IEEE Transactions on Intelligent Transportation Systems · doi:10.1109/tits.2025.3532803 · Cited by 77
“There are inherently many unknown classes in the unknown world, but it can be reduced by active learning [62], [63] and other measures.”
Semantic Segmentation with Active Semi-Supervised Learning
Rangnekar, A., Kanan, C., & Hoffman, M. J. (2022) · IEEE Workshop/Winter Conference on Applications of Computer Vision · doi:10.1109/wacv56688.2023.00591 · Cited by 44
“Query-computing methods are further classified into applying Bayesian inference [37,28,45,24,48,76,32], ensemble learning [6,68,33,63,18], diversity learning [72,85,21,77,22,53,11] and representation learning [55,41,1,42,12,19,29,30,84,87,75,62,23].”
BackgroundMethodUnbiased Scene Graph Generation in Videos
Nag, S., Min, K., Tripathi, S., & Roy–Chowdhury, A. K. (2023) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr52729.2023.02184 · Cited by 45
“The class-specific aleatoric and epistemic uncertainty, for the sample z i are computed as follows: Therefore, by using a GMM head, we are modeling the inherent uncertainty associated with the data from a Bayesian perspective [8, 9, 28].”
BackgroundMethod
Metadata
- Authors
- Choi, J., Elezi, I., Lee, H.-J., Farabet, C., & Alvarez, J. M.
- Affiliations
- Seoul National University (KR)
- Ca' Foscari University of Venice (IT)
- NVIDIA
- Venue
- 2021 IEEE/CVF International Conference on Computer Vision (ICCV)
- Publisher
- IEEE
- Published
- 2021-10
- Pages
- 10244-10253
- OpenAlex
- W3145121851
- Cited by
- 135
- References
- 49
- Citations by year
- 2026: 15 · 2025: 27 · 2024: 36 · 2023: 36 · 2022: 17 · 2021: 4
- FWCI
- 10.74
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- IEEE/CVF International Conference on Computer Vision (ICCV) (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning and Algorithms
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
- Keywords
- active learning (machine learning), object detection, mixture density networks, uncertainty estimation, aleatoric uncertainty, epistemic uncertainty, annotation cost reduction
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.
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
