PCL: Proposal Cluster Learning for Weakly Supervised Object Detection
Open accessThe DOI matches its Crossref record; 65 of the 66 references it cites were found in a registry or an index, and 1 are not checked yet.
Abstract
Weakly Supervised Object Detection (WSOD), using only image-level annotations to train object detectors, is of growing importance in object recognition. In this paper, we propose a novel deep network for WSOD. Unlike previous networks that transfer the object detection problem to an image classification problem using Multiple Instance Learning (MIL), our strategy generates proposal clusters to learn refined instance classifiers by an iterative process. The proposals in the same cluster are spatially adjacent and associated with the same object. This prevents the network from concentrating too much on parts of objects instead of whole objects. We first show that instances can be assigned object or background labels directly based on proposal clusters for instance classifier refinement, and then show that treating each cluster as a small new bag yields fewer ambiguities than the directly assigning label method. The iterative instance classifier refinement is implemented online using multiple streams in convolutional neural networks, where the first is an MIL network and the others are for instance classifier refinement supervised by the preceding one. Experiments are conducted on the PASCAL VOC, ImageNet detection, and MS-COCO benchmarks for WSOD. Results show that our method outperforms the previous state of the art significantly.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
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?
In the citing paper
“Multi-instance learning is a group of supervised learning algorithms that has seen widespread application in WSOD [202-209].”
From Object Detection in 20 Years: A Survey (2023) · cites this paper as [207]
Citations
Cites
Very 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 4×
“Here we train a Fast R-CNN (FRCNN) [7] using the VGG16 model and the same five image scales (horizontal flips only in training).”
MethodInfluentialImageNet 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 · cited 1×
“Recently, due to the development of Convolutional Neural Network (CNN) [1], [2] and the availability of large scale datasets with detailed boundingbox-level annotations [3], [4], [5], there have been great leap forwards in object detection [6], [7], [8], [9], [10], [11].”
BackgroundFaster 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 · cited 1×
“Recently, due to the development of Convolutional Neural Network (CNN) [1], [2] and the availability of large scale datasets with detailed boundingbox-level annotations [3], [4], [5], there have been great leap forwards in object detection [6], [7], [8], [9], [10], [11].”
BackgroundScikit-learn: Machine Learning in Python
Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., Blondel, M., Müller, A., Nothman, J., Louppe, G., et al. (2012) · arXiv · doi:10.48550/arxiv.1201.0490 · Cited by 71,016 · cited 1×
“The k-means algorithm to produce top ranking proposals is implemented by scikit-learn [56].”
MethodFaster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
Ren, S., He, K., Girshick, R., & Sun, J. (2015) · arXiv · doi:10.48550/arxiv.1506.01497 · Cited by 70,812 · cited 1×
“Recently, due to the development of Convolutional Neural Network (CNN) [1], [2] and the availability of large scale datasets with detailed boundingbox-level annotations [3], [4], [5], there have been great leap forwards in object detection [6], [7], [8], [9], [10], [11].”
BackgroundGradient-based learning applied to document recognition
Lecun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998) · Proceedings of the IEEE · doi:10.1109/5.726791 · Cited by 58,805 · cited 14×
“We finally visualize some success and failure detection results on VOC 2007 trainval by PCL-Ens.+FRCNN, as in Fig.”
BackgroundInfluentialYou 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 · cited 1×
“Recently, due to the development of Convolutional Neural Network (CNN) [1], [2] and the availability of large scale datasets with detailed boundingbox-level annotations [3], [4], [5], there have been great leap forwards in object detection [6], [7], [8], [9], [10], [11].”
BackgroundMicrosoft 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
ImageNet Large Scale Visual Recognition Challenge
Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., et al. (2015) · International Journal of Computer Vision · doi:10.1007/s11263-015-0816-y · Cited by 39,380
(2016) · Cited by 31,590
Self-citation
Cited by
Object Detection in 20 Years: A Survey
Zou, Z., Chen, K., Shi, Z., Guo, Y., & Ye, J. (2023) · Proceedings of the IEEE · doi:10.1109/jproc.2023.3238524 · Cited by 3,689
“Multi-instance learning is a group of supervised learning algorithms that has seen widespread application in WSOD [202], [203], [204], [205], [206], [207], [208], [209].”
A Survey of Deep Learning-Based Object Detection
Jiao, L., Zhang, F., Liu, F., Yang, S., Li, L., Feng, Z., & Qu, R. (2019) · IEEE Access · doi:10.1109/access.2019.2939201 · Cited by 1,323
“[229] utilize an iterative procedure to generate proposal clusters and learn refined instance classifiers, which makes the network concentrate on the whole object rather than part of it.”
MethodDetecting Twenty-Thousand Classes Using Image-Level Supervision
Zhou, X., Girdhar, R., Joulin, A., Krähenbühl, P., & Misra, I. (2022) · Lecture Notes in Computer Science · doi:10.1007/978-3-031-20077-9_21 · Cited by 564
Gao, Y., Ma, J., Zhao, M., Liu, W., & Yuille, A. L. (2019) · 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2019.00332 · Cited by 284
Self-citationWeakly Supervised Object Localization and Detection: A Survey
Zhang, D., Han, J., Cheng, G., & Yang, M.-H. (2021) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2021.3074313 · Cited by 277
Instance-Aware, Context-Focused, and Memory-Efficient Weakly Supervised Object Detection
Ren, Z., Yu, Z., Yang, X., Liu, M.-Y., Lee, Y. J., Schwing, A. G., & Kautz, J. (2020) · 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr42600.2020.01061 · Cited by 225
Deep Learning-Based Object Detection Techniques for Remote Sensing Images: A Survey
Li, Z., Wang, Y., Zhang, N., Zhang, Y., Zhao, Z., Xu, D., Ben, G., & Gao, Y. (2022) · Remote Sensing · doi:10.3390/rs14102385 · Cited by 206
Localizing Objects with Self-Supervised Transformers and no Labels
Siméoni, O., Puy, G., Vo, H. V., Roburin, S., Gidaris, S., Bursuc, A., Pérez, P., Marlet, R., & Ponce, J. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2109.14279 · Cited by 105
“Weakly-supervised object detection (WSOD) uses image-level labels without bounding boxes [9, 59] to learn to detect objects.”
BackgroundMethodInfluentialWang, Q., Huang, W., Xiong, Z., & Li, X. (2022) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2020.3042276 · Cited by 185
“The clustering learning technology is also attempted in [44].”
Semi-Supervised 3D Abdominal Multi-Organ Segmentation Via Deep Multi-Planar Co-Training
Zhou, Y., Wang, Y., Tang, P., Bai, S., Shen, W., Fishman, E. K., & Yuille, A. (2019) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv.2019.00020 · Cited by 211
“With the recent advance of deep learning and its applications [23, 38, 37, 39], fully convolutional networks (FCNs) [24] have been successfully applied to many biomedical segmentation tasks such as neuronal structures segmentation [9, 13, 30, 35], single organ segmentation [32, 48, 47], and multi-organ segmentation [3…”
MethodSelf-citationYao, X., Feng, X., Han, J., Cheng, G., & Guo, L. (2021) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2020.2991407 · Cited by 169
Zeng, Z., Liu, B., Fu, J., Chao, H., & Zhang, L. (2019) · 2019 IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2019.00838 · Cited by 163
“We can find that WSOD(2) significantly surpasses outperforms other end-toend trainable models [28, 29, 35] on both PASCAL VOC 2007 and 2012.”
BackgroundMethodInfluentialShao, Z., Cheng, G., Ma, J., Wang, Z., Wang, J., & Li, D. (2021) · IEEE Transactions on Multimedia · doi:10.1109/tmm.2021.3075566 · Cited by 167
“One type is two-stage methods [7]–[12], which divide detection into two parts, region proposal and classification.”
BackgroundMethodWeakly supervised mitosis detection in breast histopathology images using concentric loss
Li, C., Wang, X., Liu, W., Latecki, L. J., Wang, B., & Huang, J. (2019) · Medical Image Analysis · doi:10.1016/j.media.2019.01.013 · Cited by 160
Self-citationTowards Precise End-to-End Weakly Supervised Object Detection Network
Yang, K., Li, D., & Dou, Y. (2019) · 2019 IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2019.00846 · Cited by 133
“We follow the MIL pipeline and combine the two-stream WSDDN [1] and OICR/PCL algorithms [34, 33] to implement our basic MIL branch and refine the detected boxes with a regression branch in an online manner.”
BackgroundMethodInfluentialHigh-Quality Proposals for Weakly Supervised Object Detection
Cheng, G., Yang, J., Gao, D., Guo, L., & Han, J. (2020) · IEEE Transactions on Image Processing · doi:10.1109/tip.2020.2987161 · Cited by 153
“They are WSDDN [1], OICR [2], DSTL [55], WCCN [56], PCL [57], TS2C [43], MELM [3], ZLDN [58], C-WSL [59], WSRPN [60], ML-LocNet [61], W2F [39], C-MIL [36], Pred Net [40], OICR + GAM + REG [37], and WS-JDS [41].”
MethodWeakly Supervised Object Detection Using Proposal- and Semantic-Level Relationships
Zhang, D., Zeng, W., Yao, J., & Han, J. (2022) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2020.3046647 · Cited by 132
Learning Open-World Object Proposals Without Learning to Classify
Kim, D., Lin, T.-Y., Angelova, A., Kweon, I. S., & Kuo, W. (2022) · IEEE Robotics and Automation Letters · doi:10.1109/lra.2022.3146922 · Cited by 121
“They have become the integral pre-processing steps for many computer vision systems, including object detection [33,41,42,45,59,74], segmentation [3,8,13], object discovery [12, 16, 47], weakly supervised object detection [5, 21, 53], visual tracking [35, 63], content-aware retargeting [51], etc.”
BackgroundCyclic Guidance for Weakly Supervised Joint Detection and Segmentation
Shen, Y., Ji, R., Wang, Y., Wu, Y., & Cao, L. (2019) · 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2019.00079 · Cited by 122
“The evaluation of object detection is performed on PASCAL VOC 2007, 2010 and 2012 [20], with comparison with several state-of-the-art methods [23, 70, 68, 77, 72, 82, 24, 58].”
MethodLiu, X., Miao, X., Jiang, H., & Chen, J. (2020) · Annual Reviews in Control · doi:10.1016/j.arcontrol.2020.09.002 · Cited by 140
RefineDet++: Single-Shot Refinement Neural Network for Object Detection
Zhang, S., Wen, L., Lei, Z., & Li, S. Z. (2020) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2020.2986402 · Cited by 134
“There are some related topics to fully-supervised object detection, such as weakly-supervised object detection [57], [58] that uses only image-level annotations to train object detectors via a three step pipeline, instance segmentation [16], [59], [60] that requires the correct detection of all…”
BackgroundA Self-Reasoning Framework for Anomaly Detection Using Video-Level Labels
Zaheer, M. Z., Mahmood, A., Shin, H., & Lee, S. (2020) · IEEE Signal Processing Letters · doi:10.1109/lsp.2020.3025688 · Cited by 131
“Instead of using it for pre-training [18], [19], the clustering in our method serves two purposes: 1) It helps in generating fragment-level pseudo annotations from video-level labels.”
Weakly-Supervised Spatio-Temporally Grounding Natural Sentence in Video
Chen, Z., Ma, L., Luo, W., & Wong, K.-Y. K. (2019) · Proceedings of the conference - Association for Computational Linguistics. Meeting · doi:10.18653/v1/p19-1183 · Cited by 135
“To handle the above challenges, we propose a novel model realized within the multiple instance learning framework (Karpathy and Fei-Fei, 2015; Tang et al., 2017, 2018).”
MethodFeng, X., Han, J., Yao, X., & Cheng, G. (2020) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2020.2985989 · Cited by 114
“Recently, many popular methods [11], [12], [14], [16], [18], [22] apply MIL to address the WSOD problem and achieve promising results.”
BackgroundMethodTCANet: Triple Context-Aware Network for Weakly Supervised Object Detection in Remote Sensing Images
Feng, X., Han, J., Yao, X., & Cheng, G. (2020) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2020.3030990 · Cited by 114
“To address these issues, most of the popular algorithms [22], [24], [26] motivated by OICR [23] are eager to mine the whole object by propagating image-level label from the most discriminative part to the surrounding regions that have large spatial overlap with it.”
BackgroundMethodInfluentialComprehensive Attention Self-Distillation for Weakly-Supervised Object DetectionPreprint
Huang, Z., Zou, Y., Bhagavatula, V., & Huang, D. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2010.12023 · Cited by 161
“To have a fair comparison with other methods, following [12, 7], multi-level scaling and horizontal flipping data augmentation are conducted in training.”
BackgroundMethodInfluentialGao, Y., Liu, B., Guo, N., Ye, X., Wan, F., You, H., & Fan, D. (2019) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2019.00993 · Cited by 108
“Instead of directly labeling the positive proposals to class c , we treat these positive proposals as a bag, and use a averaged MIL pooling method [30].”
MethodInfluentialOpen Vocabulary Object Detection with Pseudo Bounding-Box Labels
Gao, M., Chen, X., Niebles, J. C., Li, J., Xu, R., Liu, W., & Xiong, C. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-20080-9_16 · Cited by 105
“Weakly supervised object detectors seek to relieve such heavy human annotation burden by using image-level labels such as image-level object categories [2,29], captions [32] and object counts [7] for training.”
BackgroundSelf-citationFew-Shot Object Detection via Classification Refinement and Distractor Retreatment
Li, Y., Zhu, H., Cheng, Y., Wang, W., Teo, C. S., Xiang, C., Vadakkepat, P., & Lee, T. H. (2021) · 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/cvpr46437.2021.01514 · Cited by 117
“In view of the data scarcity of the novel classes, such extra samples help to improve the final detection performance with zero additional annotation cost [15, 14].”
BackgroundDeep Learning for Weakly-Supervised Object Detection and Localization: A Survey
Shao, F., Chen, L., Shao, J., Ji, W., Xiao, S., Ye, L., Zhuang, Y., & Xiao, J. (2022) · Neurocomputing · doi:10.1016/j.neucom.2022.01.095 · Cited by 119
A Survey of Self-Supervised and Few-Shot Object Detection
Huang, G., Laradji, I., Vazquez, D., Lacoste-Julien, S., & Rodriguez, P. (2022) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2022.3199617 · Cited by 100
“[103] and could therefore benefit from a larger pool of labels.”
BackgroundMin-Entropy Latent Model for Weakly Supervised Object Detection
Wan, F., Wei, P., Han, Z., Jiao, J., & Ye, Q. (2019) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2019.2898858 · Cited by 87
“The iterative generation of the proposal clusters [43] with OICR prevented the network from concentrating on parts of objects.”
MethodObject Instance Mining for Weakly Supervised Object Detection
Lin, C., Wang, S., Xu, D., Lu, Y. F., & Zhang, W. (2020) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v34i07.6813 · Cited by 90
Weakly Supervised Temporal Adjacent Network for Language Grounding
Wang, Y., Deng, J., Zhou, W., & Li, H. (2021) · IEEE Transactions on Multimedia · doi:10.1109/tmm.2021.3096087 · Cited by 100
SLV: Spatial Likelihood Voting for Weakly Supervised Object Detection
Chen, Z., Fu, Z., Jiang, R., Chen, Y., & Hua, X. (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/cvpr42600.2020.01301 · Cited by 85
Point-to-Box Network for Accurate Object Detection via Single Point Supervision
Chen, P., Yu, X., Han, X., Hassan, N., Wang, K., Li, J., Zhao, J., Shi, H., Han, Z., & Ye, Q. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-20077-9_4 · Cited by 98
Unsupervised deep clustering via contractive feature representation and focal loss
Cai, J., Wang, S., Xu, C., & Guo, W. (2021) · Pattern Recognition · doi:10.1016/j.patcog.2021.108386 · Cited by 87
Wu, Z., Liu, C., Wen, J., Xu, Y. D., Yang, J. J., & Li, X. (2022) · IEEE Transactions on Image Processing · doi:10.1109/tip.2022.3231744 · Cited by 83
“Since only some works have reported their results, here we compare with 14 methods, including PCL [23], MELM [36], CSC [45], WS-JDS [34], C-MIDN [30], WSOD 2 [19], PG-PS [24], PSLR [27], MIST [33], P-MIDN + MGSC [31], GradingNet-C-MIL [28], D-MIL [32], OICR [16] and OICR + Reg.”
BackgroundMethodInfluentialWeakly Supervised 3D Object Detection from Point Clouds
Qin, Z., Wang, J., & Lu, Y. (2020) · ACM International Conference on Multimedia (ACM MM) · doi:10.1145/3394171.3413805 · Cited by 88
“These metrics have been widely adopted in previous works [11, 30, 31, 38, 39].”
MethodResultInfluentialNormality Guided Multiple Instance Learning for Weakly Supervised Video Anomaly Detection
Park, S., Kim, H., Kim, M., Kim, D., & Sohn, K. (2023) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv56688.2023.00269 · Cited by 91
“We can simply represent anomaly score a ( f i , P ) as an inverted similarity score between the instance and normality prototype such that: Finally, we refine the score by ensembling two prediction scores from each classifier, following [36, 35, 16, 49] :”
MethodGe, C., Wang, J., Wang, J., Qi, Q., Sun, H., & Liao, J. (2020) · Computers in Industry · doi:10.1016/j.compind.2020.103232 · Cited by 74
Structure-Texture Aware Network for Low-Light Image Enhancement
Xu, K., Chen, H., Xu, C., Jin, Y., & Zhu, C. (2022) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2022.3141578 · Cited by 78
Qian, X., Huo, Y., Cheng, G., Gao, C., Yao, X., & Wang, W. (2023) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2023.3266838 · Cited by 66
“Tang et al. [30] proposed a proposal cluster learning (PCL) strategy, which divided proposals into several clusters to mine multiple PGT instances.”
BackgroundMethodStrengthen Learning Tolerance for Weakly Supervised Object Localization
Guo, G., Han, J., Wan, F., & Zhang, D. (2021) · 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/cvpr46437.2021.00732 · Cited by 77
“Most methods handle WSOD as a multiple instance learning problem, in which object proposals must be provided and the model is trained to select the the most con-fident proposal [2, 33, 32, 37, 35, 23].”
MethodProposalCLIP: Unsupervised Open-Category Object Proposal Generation via Exploiting CLIP Cues
Shi, H., Hayat, M., Wu, Y., & Cai, J. (2022) · 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/cvpr52688.2022.00939 · Cited by 90
“Graph-based technologies are also used in several works [15, 39] to extract seeds and cluster centers for better proposal generation.”
MethodPointOBB: Learning Oriented Object Detection via Single Point Supervision
Luo, J., Yang, X., Yu, Y., Li, Q., Yan, J., & Li, Y. (2024) · 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/cvpr52733.2024.01583 · Cited by 89
“As to reducing the annotation cost, weakly-supervised horizontal object detection using image-level annotations has been well-developed [2, 6, 13, 14, 26, 37, 38, 41], but these methods have limited performance in complex aerial scenes and cannot predict the orientation of objects.”
BackgroundMultiple Instance Graph Learning for Weakly Supervised Remote Sensing Object Detection
Wang, B., Zhao, Y., & Li, X. (2021) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2021.3123231 · Cited by 85
Group R-CNN for Weakly Semi-supervised Object Detection with Points
Zhang, S., Yu, Z., Liu, L., Wang, X., Zhou, A., & Chen, K. (2022) · 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/cvpr52688.2022.00920 · Cited by 76
“In contrast, Group R-CNN selects proposals around the human-labeled points to improve the recall rate, which is different from PCL in terms of motivation and implementation.”
BackgroundSelf-citationZhang, D., Han, J., Zhao, L., & Zhao, T. (2020) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2020.2969483 · Cited by 62
“When compared to Tang’s (L) [54], our approach obtains slightly worse performance on the trainval data sets but better performance on the test data sets.”
ResultParameterized Hamiltonian Learning With Quantum Circuit
Shi, J., Wang, W., Lou, X., Zhang, S., & Li, X. (2022) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2022.3203157 · Cited by 75
Detect or Track: Towards Cost-Effective Video Object Detection/Tracking
Luo, H., Xie, W., Wang, X., & Zeng, W. (2019) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v33i01.33018803 · Cited by 61
Self-citationXu, Y., Zhou, C., Yu, X., Xiao, B., & Yang, Y. (2021) · IEEE Transactions on Image Processing · doi:10.1109/tip.2021.3056887 · Cited by 65
“Since an MIDN is trained with only image-level supervision and no additional constraints are imposed, it tends to detect local discriminative regions ( e.g. heads of people and animals) rather than whole objects, as pointed out in [7], [43].”
BackgroundMethodSAL:Selection and Attention Losses for Weakly Supervised Semantic Segmentation
Zhou, L., Gong, C., Liu, Z., & Fu, K. (2020) · IEEE Transactions on Multimedia · doi:10.1109/tmm.2020.2991592 · Cited by 63
“recently [1], [2], [3], [4], [5], where the goal is to assign a semantic label to a pixel.”
BackgroundUFO$$^2$$: A Unified Framework Towards Omni-supervised Object Detection
Ren, Z., Yu, Z., Yang, X., Liu, M.-Y., Schwing, A. G., & Kautz, J. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-58529-7_18 · Cited by 75
Toward Unsupervised, Multi-object Discovery in Large-Scale Image Collections
Vo, H. V., Pérez, P., & Ponce, J. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-58592-1_46 · Cited by 92
Self-supervised object detection from audio-visual correspondence
Afouras, T., Asano, Y. M., Fagan, F., Vedaldi, A., & Metze, F. (2022) · 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/cvpr52688.2022.01032 · Cited by 69
“For weakly-supervised detection, we consider PCL [89], the strongest such baseline for which we could find an implementation.”
MethodBoosting Weakly Supervised Object Detection via Learning Bounding Box Adjusters
Dong, B., Huang, Z., Guo, Y., Wang, Q., Niu, Z., & Zuo, W. (2021) · IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv48922.2021.00287 · Cited by 68
“Existing WSOD methods [3, 25, 34, 20] usually adopt the multiple instance learning (MIL) framework based on the precomputed proposals.”
MethodResultInfluentialWeakly Supervised Rotation-Invariant Aerial Object Detection Network
Feng, X., Yao, X., Cheng, G., & Han, J. (2022) · 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/cvpr52688.2022.01375 · Cited by 57
“To disclose the contribution of the proposed instance mining strategy, we first integrate proposal cluster learning [30] into our RINet to mine the same class instances.”
BackgroundMethodInfluentialXu, Y., Sun, Y., Yang, Z., Miao, J., & Yang, Y. (2022) · 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/cvpr52688.2022.01393 · Cited by 80
“Most prior weakly supervised object detection (WSOD) arts [3, 12, 29, 31, 46, 53, 54, 58, 62, 63, 73, 75, 77] focus on learning object detectors with only image-level annotations.”
BackgroundMethodDiscrepant multiple instance learning for weakly supervised object detection
Gao, W., Wan, F., Yue, J., Xu, S., & Ye, Q. (2021) · Pattern Recognition · doi:10.1016/j.patcog.2021.108233 · Cited by 58
Product1M: Towards Weakly Supervised Instance-Level Product Retrieval via Cross-Modal Pretraining
Zhan, X., Wu, Y., Dong, X., Wei, Y., Lu, M., Zhang, Y., Xu, H., & Liang, X. (2021) · 2021 IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv48922.2021.01157 · Cited by 54
“PCL [39] iteratively generates proposal clusters to facilitate the learning of instance classifiers.”
Contextual Adversarial Attacks For Object Detection
Zhang, H., Zhou, W., & Li, H. (2020) · IEEE International Conference on Multimedia and Expo (ICME) · doi:10.1109/icme46284.2020.9102805 · Cited by 63
“Typical frameworks like OICR [25] and PCL [26] design recursive refinement branches to provide pseudo-labels.”
MethodIqbal, J., Munir, M. A., Mahmood, A., Ali, A. R., & Ali, M. (2021) · Neurocomputing · doi:10.1016/j.neucom.2021.01.075 · Cited by 59
“Based on [37], another work proposed is proposal cluster learning using image level annotations for object detection [38].”
BackgroundEnabling Deep Residual Networks for Weakly Supervised Object Detection
Shen, Y., Ji, R., Wang, Y., Chen, Z., Zheng, F., Huang, F., & Wu, Y. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-58598-3_8 · Cited by 59
“The m AP follows standard PASCAL VOC protocol to report the m AP at 50% Intersection-over-Union (IoU) of the detected boxes replacing the backbones of OICR [56], PCL [55] and C-MIL [59] with ResNet-WS sets the new state-of-the-art results with improvements of 10 .”
BackgroundMethodInfluentialLearning Efficient and Adaptive Cross-Channel Dependencies for Weakly-Supervised Object Detection
Chen, P., Nie, X., Ning, Y., & Zhang, Y. (2025) · IEEE Transactions on Multimedia · doi:10.1109/tmm.2025.3618564 · Cited by 53
“To further enhance the performance of weakly supervised object detectors, recent works have proposed numerous improvements to the self-training process [7], [11], and feature representation learning [12], [14].”
A Real-Time Action Representation With Temporal Encoding and Deep Compression
Liu, K., Liu, W., Ma, H., Tan, M., & Gan, C. (2020) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2020.2984569 · Cited by 54
“More specially, T-C3D improves the very recent method in [42] from to 76.3%, which adopts the multiple instance learning algorithm [51], [52] to detect anomaly in surveillance videos.”
MethodInfluentialSelf-citationPEN: Pose-Embedding Network for Pedestrian Detection
Jiao, Y., Yao, H., & Xu, C. (2020) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2020.3000223 · Cited by 52
“For weakly supervised detectors, Tang et al. [14] generate proposal clusters to learn refined instance classifiers by an iterative process with only visual information.”
BackgroundOmni-DETR: Omni-Supervised Object Detection with Transformers
Wang, P., Cai, Z., Yang, H., Swaminathan, G., Vasconcelos, N. M., Schiele, B., & Soatto, S. (2022) · 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/cvpr52688.2022.00915 · Cited by 66
“This establishes a much stronger baseline than any previous weakly supervised object detection (WSOD) and weakly semi-supervised object detection (WSSOD) work [13, 23, 32, 44].”
InfluentialRecent Few-shot Object Detection Algorithms: A Survey with Performance Comparison
Liu, T., Zhang, L., Wang, Y., Guan, J., Fu, Y., Zhao, J., & Zhou, S. (2023) · ACM Transactions on Intelligent Systems and Technology · doi:10.1145/3593588 · Cited by 55
“Semi-supervised object detection [22, 36, 61, 131, 132, 147] and weakly-supervised object detection [4, 5, 23, 118, 130, 139, 180] have been developed for a long time, but most of these methods still require much more data than FSOD methods.”
MethodCap2Det: Learning to Amplify Weak Caption Supervision for Object Detection
Ye, K., Zhang, M., Kovashka, A., Li, W., Qin, D., & Berent, J. (2019) · 2019 IEEE/CVF International Conference on Computer Vision (ICCV) · doi:10.1109/iccv.2019.00978 · Cited by 46
“We compare to a representative fully-supervised detection model [26]; “Faster Inception-V2” [15] which is our method’s supervised detection counterpart, and a recent WSOD model [30].”
BackgroundMethodInfluentialMonoGRNet: A General Framework for Monocular 3D Object Detection
Qin, Z., Wang, J., & Lu, Y. (2021) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2021.3074363 · Cited by 53
“[54] localizes objects by clustering.”
BackgroundEnhanced Spatial Feature Learning for Weakly Supervised Object Detection
Wu, Z., Wen, J., Xu, Y. D., Yang, J. J., Li, X., & Zhang, D. D. (2022) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2022.3178180 · Cited by 45
“They further improve OICR by generating proposal clusters and their spatial adjacent proposals to train refined instance classifiers [20].”
BackgroundMethodWhat Can Human Sketches Do for Object Detection?
Chowdhury, P. N., Bhunia, A. K., Sain, A., Koley, S., Xiang, T., & Song, Y.-Z. (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/cvpr52729.2023.01448 · Cited by 61
“PCL [83] generates multiple positive instance in an image via clustering and assigning proposals to the label of corresponding object class for each cluster.”
BackgroundZhao, R., Liao, W., Zou, B., Chen, Z., & Li, S. (2019) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v33i01.3301809 · Cited by 45
“Recently, many methods are emerging to establish the desired pixel-label correspondence in training for weaklysupervised learning (Tang et al. 2018; Huang et al. 2018; Kwak et al. 2017).”
MethodScalable logo detection by self co-learning
Su, H., Gong, S., & Zhu, X. (2019) · Pattern Recognition · doi:10.1016/j.patcog.2019.107003 · Cited by 44
“SSD [18] 8.8 Faster R-CNN [4] 14.9 YOLOv2 [5] 18.4 YOLOv3 [20] 11.0 RetinaNet [19] 4.1 WSL [21] 3.6 PCL [22] 0.2 WLOD [23] 19.3 WLOD [23] + SCL [7] 7.8 ULD [14,29] 13.2 SLST [24] 36.8 SL 2 (Ours) 46.9 procedure, we propose to incrementally enhance the model capability in a joint spirit of…”
InfluentialInstance Mining with Class Feature Banks for Weakly Supervised Object Detection
Yin, Y., Deng, J., Zhou, W., & Li, H. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i4.16429 · Cited by 45
“PCL (Tang et al. 2018) 40”
InfluentialObject Discovery via Contrastive Learning for Weakly Supervised Object Detection
Seo, J., Bae, W., Sutherland, D. J., Noh, J., & Kim, D. (2022) · Lecture notes in computer science · doi:10.1007/978-3-031-19821-2_18 · Cited by 60
Boosting Weakly Supervised Object Detection with Progressive Knowledge Transfer
Zhong, Y., Wang, J., Peng, J., & Zhang, L. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-58574-7_37 · Cited by 59
Dual Attention Networks for Few-Shot Fine-Grained Recognition
Xu, S.-L., Zhang, F., Wei, X.-S., & Wang, J. (2022) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v36i3.20196 · Cited by 60
“In particular, the MIL methods for image recognition (Tang et al. 2018; Carbonneau et al. 2018; Angles et al. 2021) has also achieved some good results.”
MethodSelf-Guided Proposal Generation for Weakly Supervised Object Detection
Cheng, G., Xie, X., Chen, W., Feng, X., Yao, X., & Han, J. (2022) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2022.3181466 · Cited by 45
“In particular, our SPG method (with MELM) brings about the mAP improvements of 27.68 % , 28.28 % , 20.51 % and 23.39 % , compared with WSDDN [25], OICR [26], MELM [27] and PCL [28], respectively.”
BackgroundMethodResultInfluentialMissing Labels in Object Detection
Xu, M., Bai, Y., & Ghanem, B. (2019) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · Cited by 59
Feng, X., Yao, X., Cheng, G., Han, J., & Han, J. (2021) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2021.3105575 · Cited by 44
“Based on it, some researches aim to leverage curriculum learning [32], better initialization models [23], [27], knowledge distillation [24], [33], and regularization [34], [35] to further improve the performance of WSOD.”
BackgroundInfluentialObject Localization under Single Coarse Point Supervision
Yu, X., Chen, P., Wu, D., Hassan, N., Li, G., Yan, J., Shi, H., Ye, Q., & Han, Z. (2022) · 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/cvpr52688.2022.00482 · Cited by 47
“MIL is widely used in weakly supervised object localization and detection (WSOL [8, 38] and WSOD [4, 7, 32, 34–36]), where only the image-level annotation is utilized.”
MethodWang, G., Zhang, X., Peng, Z., Jia, X., Tang, X., & Jiao, L. (2023) · ISPRS Journal of Photogrammetry and Remote Sensing · doi:10.1016/j.isprsjprs.2023.01.011 · Cited by 42
Khan, H. U., Raza, B., Shah, M. H., Usama, S. M., Tiwari, P., & Band, S. S. (2022) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2022.104414 · Cited by 40
WSODet: A Weakly Supervised Oriented Detector for Aerial Object Detection
Tan, Z., Jiang, Z., Guo, C., & Zhang, H. (2023) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2023.3247578 · Cited by 52
“1, the procedure of the mainstream WSOD methods [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19] is as follows: a large number of proposals are obtained by proposal extraction algorithms before training and testing.”
BackgroundMethodInfluentialQian, X., Zhang, B., He, Z., Wang, W., Yao, X., & Cheng, G. (2025) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2025.3586239 · Cited by 39
“MIST [41] chooses the proposals with top-K highest CCSs as seed instances instead of the proposals with the highest CCS. PCL [42] divides proposals into clusters based on their CCSs, and selects the cluster with the highest CCS as seed instances.”
Object Discovery From a Single Unlabeled Image by Mining Frequent Itemsets With Multi-Scale Features
Zhang, R., Huang, Y., Pu, M., Zhang, J., Guan, Q., Zou, Q., & Ling, H. (2020) · IEEE Transactions on Image Processing · doi:10.1109/tip.2020.3015543 · Cited by 48
“However, for the challenging PAS-CAL VOC dataset, our proposed OLM is still far behind the weakly object localization methods, e.g., Proposal Cluster Learning (PCL) method [83] currently achieves a promising CorLoc of 66.6% on trainval set.”
MethodInfluentialDeep graph cut network for weakly-supervised semantic segmentation
Feng, J., Wang, X., & Liu, W. (2021) · Science China Information Sciences · doi:10.1007/s11432-020-3065-4 · Cited by 40
Self-citationDynamic Pseudo-Label Generation for Weakly Supervised Object Detection in Remote Sensing Images
Wang, H., Li, H., Qian, W., Diao, W., Zhao, L., Zhang, J., & Zhang, D. (2021) · Remote Sensing · doi:10.3390/rs13081461 · Cited by 39
“[19] proposed proposal cluster learning (PCL) to optimize the process of online instance classifier refinement.”
MethodInfluentialMultiple Instance Detection Networks With Adaptive Instance Refinement
Wu, Z., Wen, J., Xu, Y. D., Yang, J. J., & Zhang, D. D. (2021) · IEEE Transactions on Multimedia · doi:10.1109/tmm.2021.3125130 · Cited by 39
“Following [29], [45], [47], we train our model on the train set (82,783 images) and test on the val set (40,504 images).”
BackgroundMethodInfluentialQian, X., Li, C., Wang, W., Yao, X., & Cheng, G. (2023) · International Journal of Applied Earth Observation and Geoinformation · doi:10.1016/j.jag.2023.103301 · Cited by 44
Qian, X., Wang, C., Wang, W., Yao, X., & Cheng, G. (2024) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2024.3408895 · Cited by 37
“Other congener methods include proposal cluster learning (PCL) [32], weakly supervised contrastive loss (WSCL) [22], and so on.”
BackgroundMethodInfluentialContext-aware co-supervision for accurate object detection
Peng, J., Wang, H., Yue, S., & Zhang, Z. (2021) · Pattern Recognition · doi:10.1016/j.patcog.2021.108199 · Cited by 38
Environment Perception Technologies for Power Transmission Line Inspection Robots
Chen, M., Tian, Y., Xing, S., Li, Z., Li, E., Liang, Z., & Guo, R. (2021) · Journal of Sensors · doi:10.1155/2021/5559231 · Cited by 35
“Consequently, researchers try to employ weakly supervised data to train models [112, 113].”
BackgroundWang, Y., Tang, P., Zhou, Y., Shen, W., Fishman, E. K., & Yuille, A. L. (2021) · IEEE Transactions on Medical Imaging · doi:10.1109/tmi.2021.3060066 · Cited by 34
Self-citationSemantic-aware SAM for Point-Prompted Instance Segmentation
Wei, Z., Chen, P., Yu, X., Li, G., Jia, J., & Han, Z. (2024) · 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/cvpr52733.2024.00344 · Cited by 55
“Inspired by PCL [38], we consider refining the proposals in a second phase.”
MethodAccurate Pedestrian Detection by Human Pose Regression
Zhao, Y., Yuan, Z., & Chen, B. (2019) · IEEE Transactions on Image Processing · doi:10.1109/tip.2019.2942686 · Cited by 38
“Driven by the success of CNNs on object detection [39], [45]–[47], we fuse our method with detectors using CNNs.”
MethodModeling Sub-Actions for Weakly Supervised Temporal Action Localization
Huang, L., Huang, Y., Ouyang, W., & Wang, L. (2021) · IEEE Transactions on Image Processing · doi:10.1109/tip.2021.3078324 · Cited by 34
Li, Y., Li, X., Zhang, Y., Peng, D., & Bruzzone, L. (2023) · International Journal of Applied Earth Observation and Geoinformation · doi:10.1016/j.jag.2023.103345 · Cited by 46
Metadata
- Authors
- Tang, P., Wang, X., Bai, S., Shen, W., Bai, X., Liu, W., & Yuille, A.
- Affiliations
- Huazhong University of Science and Technology (CN)
- University of Oxford (GB)
- Shanghai University (CN)
- Johns Hopkins University (US)
- Venue
- IEEE Transactions on Pattern Analysis and Machine Intelligence
- Published
- 2020-01-01
- Volume
- 42
- Issue
- 1
- Pages
- 176-191
- ISSN
- 0162-8828 (ISSN-L)
- 2160-9292
- 1939-3539
- OpenAlex
- W2813911573
- Cited by
- 406
- References
- 62
- Citations by year
- 2026: 16 · 2025: 41 · 2024: 54 · 2023: 59 · 2022: 74 · 2021: 82 · 2020: 50
- FWCI
- 20.27
- Type
- journal article
- Language
- en
- Open access
- green
- arxiv.org
- Licence
- Locations
- IEEE Transactions on Pattern Analysis and Machine Intelligence (publishedVersion)
- PubMed (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- arXiv (Cornell University)
- Funders
- National Natural Science Foundation of China (61733007, 61572207, 61876212, 61672336)
- ONR (N00014-15-1-2356)
- Hubei Scientific and Technical Innovation Key Project
- HUST Academic Frontier Youth Team
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Neural Network Applications
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Image and Video Retrieval Techniques
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Visual Attention and Saliency Detection
- Keywords
- weakly supervised object detection, multiple instance learning, image-level annotations, iterative refinement, multi-stream convolutional neural network
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
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.