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
Items 201–300 of 409
CLIP-Driven Transformer for Weakly Supervised Object Localization
Chen, Z., Shen, Y., Cao, L., Zhang, S., & Ji, R. (2025) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2025.3548704 · Cited by 16
“Most approaches [54], [55], [56], [57], [58] employ multiple-instance learning (MIL) to recast WSOD as a multi-label classification problem.”
Wang, Y., Guerrero, R., & Pavlović, V. (2023) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv56688.2023.00011 · Cited by 11
“Thus, reducing these costs is of great importance, and many weakly-supervised object detection (WSOD) methods [2, 29, 28] have been developed accordingly.”
BackgroundMethodInfluentialLL-WSOD: Weakly supervised object detection in low-light
Zhang, H., Wang, Y., & Yang, Y. (2023) · Journal of Visual Communication and Image Representation · doi:10.1016/j.jvcir.2023.104010 · Cited by 11
Weakly Supervised Few-Shot Object Detection with DETR
Zhang, C., Zhang, Y., Zhang, L., Zhao, J., Guan, J., & Zhou, S. (2024) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v38i7.28528 · Cited by 11
“Recently, MIL-based methods (Bilen and Vedaldi 2016; Tang et al. 2017, 2018; Zeng et al. 2019) have gradually become mainstream.”
MethodBoosting Weakly Supervised Object Detection using Fusion and Priors from Hallucinated Depth
Gungor, C., & Kovashka, A. (2024) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv57701.2024.00079 · Cited by 11
“OICR [35] improved upon this by proposing pseudo-ground-truth mining and an online instance refinement, which was further refined by proposal clustering [34].”
BackgroundBoosting Semi-supervised Crowd Counting with Scale-based Active Learning
Zhang, S., Ke, W., Liu, S., Hong, X., & Zhang, T. (2024) · ACM International Conference on Multimedia (ACM MM) · doi:10.1145/3664647.3680976 · Cited by 18
“4 .4.”
Weakly Supervised Group Mask Network for Object Detection
Song, L., Liu, J., Sun, M., & Shang, X. (2020) · International Journal of Computer Vision · doi:10.1007/s11263-020-01397-w · Cited by 11
“As shown in recent approaches (Ren et al. 2020; Tang et al. 2018a; Durand et al. 2017), the WSD task can be commonly formulated as a multiple instance learning (MIL) (Dietterich et al. 1997) problem, where an image is represented as a bag of instances (i.e, regions).”
MethodInfluentialMulti-peak Graph-based Multi-instance Learning for Weakly Supervised Object Detection
Ji, R., Liu, Z., Zhang, L., Liu, J., Zuo, X., Wu, Y., Zhao, C., Wang, H., & Yang, L. (2021) · ACM Transactions on Multimedia Computing Communications and Applications · doi:10.1145/3432861 · Cited by 11
“…Table 1, we can conclude that the proposed method achieves the best method improves the original OICR [72] by 9.2% and outperforms other approaches such as WSDDN [4] (by 11.2%), PCL [71] (by 6.9%), WeakRPN [74] (by 5.1%), SDCN [44] (by 2.1%), Boosted OICR [19] (by 0.7%), and WS-JDS [63] (by 4.8%).”
MethodInfluentialWSSOD: A New Pipeline for Weakly- and Semi-Supervised Object DetectionPreprint
Fang, S., Cao, Y., Wang, X., Chen, K., Lin, D., & Zhang, W. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2105.11293 · Cited by 11
“WSOD [3, 38, 22, 33, 39] avoids the high cost of labeling bounding boxes with weakly-annotated data.”
BackgroundSelf-citationBFCP: Pursue Better Forward Compatibility Pretraining for Few-Shot Class-Incremental Learning
Fu, Z., Wang, Z., Xu, X., Guo, W., Chi, Z., Yang, H., & Du, W. (2025) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2025.3548465 · Cited by 10
“Third, the cluster loss [42], [43], denoted as L Cluster , aims to group samples of the same class closer together to improve the computation of better class prototypes.”
R-CCF: region-aware continual contrastive fusion for weakly supervised object detection
Zhang, Y., Tian, R., Zhang⋆, Y., Zhang, Z., Bai, Y., Ding, M., & Zuo, W. (2024) · Applied Intelligence · doi:10.1007/s10489-024-05403-3 · Cited by 10
“…methods into the following categories:initialhigh-qualityregionproposal-basedmethods[6, 7, 38–40], refining instance classifier-based methods [3, 22, 23, 41, 42], mining completely object regions-based methods [16, 17, 43–45] and weakly-supervised to fully-supervised conversion methods [4, 5,…”
InfluentialViewpoint Recommendation Based on Object-Oriented 3D Scene Reconstruction
Li, K., Wu, Y., Xue, Y., & Qian, X. (2020) · IEEE Transactions on Multimedia · doi:10.1109/tmm.2020.2981237 · Cited by 10
Training object detectors from few weakly-labeled and many unlabeled images
Yang, Z., Shi, M., Xu, C., Ferrari, V., & Avrithis, Y. (2021) · Pattern Recognition · doi:10.1016/j.patcog.2021.108164 · Cited by 11
Dynamic sample weighting for weakly supervised object detection
Li, X., Song, Y., Zhang, R., Fu, X., Jiang, H., Wang, C., Liu, Z.-Q., Gao, J., Yu, J. X., Yu, M., et al. (2022) · Image and Vision Computing · doi:10.1016/j.imavis.2022.104444 · Cited by 11
Contrastive Proposal Extension With LSTM Network for Weakly Supervised Object Detection
Lv, P., Hu, S., & Hao, T. (2022) · IEEE Transactions on Image Processing · doi:10.1109/tip.2022.3216772 · Cited by 11
“For the purpose of detecting the complete region of objects, recent work has focused on adopting cascaded refinements of MIL classifications [7], using continuity strategies [21] and modeling uncertainty [25].”
BackgroundMethodLi, K., Shan, M., Perez, S. B., Luo, K., & Worrall, S. (2025) · IEEE Robotics and Automation Letters · doi:10.1109/lra.2025.3604697 · Cited by 9
“…address this, various robotic paradigms have been proposed to train models with limited supervision, including self-supervised learning (SSL), semi-supervised learning object detection (SSOD), and weakly supervised learning object detection (WSOD) [8], [9], [10], [11], [12], [13], [14], [15], [16].”
Three Recipes for Better 3D Pseudo-GTs of 3D Human Mesh Estimation in the Wild
Moon, G., Choi, H., Chun, S., Lee, J., & Yun, S. (2023) · IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw59228.2023.00276 · Cited by 13
“The weak supervision might make networks converge to sub-optimal points as it involves ambiguity in nature compared to the full supervision [2,4,5,9,41,43].”
Dynamic proposal sampling for weakly supervised object detection
Jiang, W., Zhao, Z., Su, F., & Fang, Y. (2021) · Neurocomputing · doi:10.1016/j.neucom.2021.02.018 · Cited by 9
Deeply Activated Salient Region for Instance Search
Xiao, H.-C., Zhao, W., Jie, L., Hong, Y.-G., & Ngo, C. (2022) · ACM Transactions on Multimedia Computing Communications and Applications · doi:10.1145/3510004 · Cited by 10
“The existing approaches include proposal clustering learning (PCL) [26], multiple instance learning (MIL) [27], [28], [29] and weakly supervised instance segmentation (WSIS) [30].”
WSRC: Weakly Supervised Faster RCNN Toward Accurate Traffic Object Detection
He, Q., Liu, J., & Huang, Z. (2022) · IEEE Access · doi:10.1109/access.2022.3231293 · Cited by 9
“Moreover, our WSRC also outperforms by a large margin when compare with the state-of-the-art methods [13], [15], [50].”
BackgroundInstance search based on weakly supervised feature learning
Lin, J., Yu, Z., & Zhao, W. (2019) · Neurocomputing · doi:10.1016/j.neucom.2019.11.029 · Cited by 9
WSODPB: Weakly supervised object detection with PCSNet and box regression module
Yi, S., Ma, H., Li, X., & Wang, Y. (2020) · Neurocomputing · doi:10.1016/j.neucom.2020.08.028 · Cited by 9
What Leads to Generalization of Object Proposals?
Wang, R., Mahajan, D. K., & Ramanathan, V. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-66096-3_32 · Cited by 13
Online Active Proposal Set Generation for weakly supervised object detection
Jin, R., Lin, G., & Wen, C. (2021) · Knowledge-Based Systems · doi:10.1016/j.knosys.2021.107726 · Cited by 9
Attentional and adversarial feature mimic for efficient object detection
Wang, H., Chen, Y., Wu, M., Zhang, X., Huang, Z., & Mao, W. (2022) · The Visual Computer · doi:10.1007/s00371-021-02363-4 · Cited by 10
Müller, P., Meissen, F., Kaissis, G., & Rueckert, D. (2025) · IEEE Transactions on Medical Imaging · doi:10.1109/tmi.2024.3435015 · Cited by 8
“…iteratively using multiple refinement streams [20], incorporating the scores of larger context around the region [5], clustering spatially adjacent regions of the same class [21], or maximizing the loss for the most discriminative region to force the model to focus on larger regions [24].”
MethodTransformer-based Dynamic Fusion Clustering Network
Zhang, C., Zhao, Y., & Wang, J. (2022) · Knowledge-Based Systems · doi:10.1016/j.knosys.2022.109984 · Cited by 11
Complete Instances Mining for Weakly Supervised Instance Segmentation
Li, Z., Zeng, Z., Liang, Y., & Yu, J.-G. (2023) · International Joint Conference on Artificial Intelligence · doi:10.24963/ijcai.2023/127 · Cited by 8
“We combine Binary Cross-Entropy and PCL [Tang et al., 2020] losses as the objective function.”
BackgroundMethodInfluentialEnhanced Attention Guided Teacher–Student Network for Weakly Supervised Object Detection
Li, M., Gao, Y., Cai, W., Yang, W., Huang, Z., Hu, X., & Leung, V. C. M. (2024) · Neurocomputing · doi:10.1016/j.neucom.2024.127910 · Cited by 8
Heterogeneity-Aware Recurrent Neural Network for Hyperspectral and Multispectral Image Fusion
Lu, R., Chen, B., Sun, J., Chen, W., Wang, P., Chen, Y., Liu, H., & Varshney, P. K. (2022) · IEEE Journal of Selected Topics in Signal Processing · doi:10.1109/jstsp.2022.3180896 · Cited by 8
PistonNet: Object Separating From Background by Attention for Weakly Supervised Ship Detection
Yang, Y. H., Pan, Z., Hu, Y., & Ding, C. (2022) · IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing · doi:10.1109/jstars.2022.3184637 · Cited by 8
Shi, P., Xin, J., Wu, J., Deng, Y., Cai, Z., Du, S., & Zheng, N. (2023) · Journal of Biophotonics · doi:10.1002/jbio.202200343 · Cited by 8
Tan, Z., Jiang, Z., Chen, G., & Zhang, H. (2023) · IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing · doi:10.1109/jstars.2023.3295732 · Cited by 8
“WSDDN [1], OICR [2], PCL [3], MELM [54], PICR [7], TCANet [8], SAENet [10], and WSODet [13] by 14.”
BackgroundMethodInfluentialDetecting Human-Object Interaction with Mixed Supervision
Kumaraswamy, S., Shi, M., & Ewa Kijak, C. (2021) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv48630.2021.00127 · Cited by 8
“Weakly-supervised learning in visual recognition is mostly focused on object detection [3, 40, 44, 39, 43, 51].”
WS-RCNN: Learning to Score Proposals for Weakly Supervised Instance Segmentation
Ou, J., Deng, S., & Yu, J.-G. (2021) · Sensors · doi:10.3390/s21103475 · Cited by 9
CAM-based non-local attention network for weakly supervised fire detection
Wang, W., Lai, L., Chen, J., & Wu, Q. (2022) · Service Oriented Computing and Applications · doi:10.1007/s11761-022-00336-6 · Cited by 8
Learning to Overcome Noise in Weak Caption Supervision for Object Detection
Unal, M. E., Ye, K., Zhang, M., Thomas, C., Kovashka, A., Li, W., Qin, D., & Berent, J. (2022) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2022.3187350 · Cited by 11
“We compare to a representative fully-supervised detection model [1]; “Faster Inception-V2” [58] which is our WSOD variant’s supervised detection counterpart (using bounding-box annotations), and a recent WSOD model, PCL-OB-G Ens + FRCNN [16].”
MethodInfluentialProposals Generation for Weakly Supervised Object Detection in Artwork Images
Milani, F., Vago, N. O. P., & Fraternali, P. (2022) · Journal of Imaging · doi:10.3390/jimaging8080215 · Cited by 9
“The proposed pipeline is compared against state-of-the-art WSOD architectures chosen based on their novelty and outstanding results on natural images: PCL [30], CASD [24] and UWSOD [36].”
BackgroundMethodCPR++: Object Localization via Single Coarse Point Supervision
Yu, X., Chen, P., Wang, K., Han, X., Li, G., Han, Z., Ye, Q., & Jiao, J. (2024) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2024.3361472 · Cited by 7
“Firstly, Select Search [50], Edge Box [51] or MCG [52] methods are used to produce proposal boxes, which are then used as a bag and each of them as an instance.”
MethodBai, R. (2025) · Knowledge-Based Systems · doi:10.1016/j.knosys.2025.113033 · Cited by 14
Denoising and Feature Enhancement Network for Target Detection Based on SAR Images
Yang, C. F., Li, C., & Zhu, Y. (2025) · Remote Sensing · doi:10.3390/rs17101739 · Cited by 7
Instance-Level Contrastive Learning for Weakly Supervised Object Detection
Zhang, M., & Zeng, B. (2022) · Sensors · doi:10.3390/s22197525 · Cited by 7
“Since Hakan Bilen and Andrea Vedaldi proposed a weakly supervised deep detection network (WSDDN) [20] to combine MIL and CNN into an end-to-end network, most WSOD methods follow [14–19, 21, 22, 26, 27 ] this pipeline to train the weakly supervised detector.”
MethodInfluentialClassifier Refinement for Weakly Supervised Object Detection with Class-Specific Activation Map
Du, P., Zhang, H., & Ma, H. (2019) · Proceedings - International Conference on Image Processing · doi:10.1109/icip.2019.8803672 · Cited by 7
“The structure of loss function is same with OICR: Lkr = − 1|R| ∑|R| r=1 ∑C+1 c=1 w k r y ck r logx ck r , where R is the total number of proposals in this image, C is the class number, k is the refinement number, and the wkr = x Rk cjkc is the loss weight.”
MethodInfluentialLearning to Detect and Retrieve Objects From Unlabeled Videos
Amrani, E., Ben‐Ari, R., Hakim, T., & Bronstein, A. M. (2019) · IEEE International Conference on Computer Vision Workshops · doi:10.1109/iccvw.2019.00567 · Cited by 7
“As our baseline, we opted for the Proposal Cluster Learning (PCL) WSOD method [15] currently achieving the highest mAP of 19.6% for WSOD on ImageNet-Det.”
MethodInfluentialSpatial likelihood voting with self-knowledge distillation for weakly supervised object detection
Chen, Z., Fu, Z., Huang, J., Tao, M., Jiang, R., Tian, X., Chen, Y., & Hua, X. (2021) · Image and Vision Computing · doi:10.1016/j.imavis.2021.104314 · Cited by 7
“The multiple instance learning (MIL) paradigm is followed in most recent WSOD studies [16, 17, 18, 19, 20, 21].”
MethodInfluentialTask-Aware Weakly Supervised Object Localization With Transformer
Meng, M., Zhang, T., Zhang, Z., Zhang, Y., & Wu, F. H. (2022) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2022.3230902 · Cited by 13
“To alleviate such a heavy burden, weakly supervised learning that uses training data with incomplete annotations has drawn increasing attention in various computer vision tasks [7], [8], [9], [10], [11].”
BackgroundCollaborative Learning-Based Network for Weakly Supervised Remote Sensing Object Detection
Chen, S., Wang, H., Mukherjee, M., & Xu, X. (2022) · IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing · doi:10.1109/jstars.2022.3223845 · Cited by 7
“pervised object detection to improve detection accuracy by using pseudo-labels output from weakly supervised networks to train strongly supervised networks [43], [44], [45], [46], [47].”
MethodEnd-to-end Point Supervised Object Detection with low-level instance features
Chen, X., Yang, C., Mo, J., Jiang, Y., & Zheng, Z. (2024) · Applied Soft Computing · doi:10.1016/j.asoc.2024.111513 · Cited by 7
Learning Object Scale With Click Supervision for Object Detection
Zhang, L., Yan, Y., Cheng, L., & Wang, H. (2019) · IEEE Signal Processing Letters · doi:10.1109/lsp.2019.2937387 · Cited by 7
“[14] 83.”
Weakly Supervised Object Detection Using Complementary Learning and Instance Clustering
Awan, M., & Shin, J. (2020) · IEEE Access · doi:10.1109/access.2020.2999596 · Cited by 7
“Complementary feature learning together with IC is a better, robust and computationally efficient approach than instance classifier refinement used in [5] and [14].”
MethodInfluentialTTFNeXt for real-time object detection
Liu, Z., Zheng, T., Xu, G., Yang, Z., Li, X., & Cai, D. (2020) · Neurocomputing · doi:10.1016/j.neucom.2020.12.055 · Cited by 7
Cross-Supervised Object DetectionPreprint
Chen, Z., Shen, Z., Yu, J., & Miller, E. G. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.15056 · Cited by 7
“Tang et al. (2017) and Tang et al. (2018) find that several iterations of online refinement based on the outputs of previous iterations boosts performance.”
BackgroundMethodInfluentialWeakly Supervised Keypoint DiscoveryPreprint
Ryou, S., & Perona, P. (2021) · CaltechAUTHORS (California Institute of Technology) · doi:10.48550/arxiv.2109.13423 · Cited by 7
Towards Open Vocabulary Object Detection without Human-provided Bounding BoxesPreprint
Gao, M., Chen, X., Niebles, J. C., Li, J., Xu, R., Liu, W., & Xiong, C. (2021) · arXiv (Cornell University) · Cited by 7
Self-citationPointSR: Self-Regularized Point Supervision for Drone-View Object Detection
Li, W., Xi, Y., Jia, W., Zhang, Z., Li, F., Liu, X., & Miao, Q. (2025) · 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/cvpr52734.2025.01093 · Cited by 6
“Inspired by this, Tang et al. [24] designed a Pro-posal Cluster Learning (PCL) approach, which used proposal clusters as supervision to infer rough object locations.”
An efficient weakly semi-supervised method for object automated annotation
Wang, X., Wei, G., Chen, S., & Liu, J. (2023) · Multimedia Tools and Applications · doi:10.1007/s11042-023-15305-0 · Cited by 7
“On the other hand, weakly supervised methods explore more time-efficient forms of weakly labeled annotations, such as class tags [18, 23, 30, 49, 54, 55], global binary image labels [42, 53], eye-tracking [36], scribbles [41], speech [20], visual marker [27, 28], geographic context information…”
Self-citationZeng, L., Huo, Y., Qian, X., & Chen, Z. (2023) · Electronics · doi:10.3390/electronics12132758 · Cited by 6
IDO: Instance dual-optimization for weakly supervised object detection
Ren, Z., Tang, Y., & Zhang, W. (2023) · Applied Intelligence · doi:10.1007/s10489-023-04956-z · Cited by 6
“Although many WSOD methods, e.g., the weakly supervised deep detection network (WSDDN) [14] and its variants [11, 15, 16], propose various refinement strategies to address the aforementioned issues, their performance leaves much to be desired.”
BackgroundUniDetector: Towards Universal Object Detection With Heterogeneous Supervision
Wang, Z., Li, Y., Chen, X., Lim, S.-N., Torralba, A. B., Zhao, H., & Wang, S. (2024) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2024.3411595 · Cited by 10
“Weakly-supervised object detection methods [46], [47], [48], [49] use only image-level labels to train object detectors, semi-supervised object detection meth-ods [50], [51], [52], [53] utilize fully-labeled images and unlabeled images for object detection, and semi-supervised weakly-supervised…”
Prompt3D: Random Prompt Assisted Weakly-Supervised 3D Object Detection
Zhang, X., Ye, H., Li, J., Tang, Q., Li, Y., Guo, Y., & Guo, J. J. (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/cvpr52733.2024.02649 · Cited by 9
“As explored in [3, 7, 10, 14, 18, 45, 49, 50], it is assumed that the instance-level bounding box annotations are not provided by the training set, and the supervision can come from weak annotations.”
Jiang, S., Liu, M., Liu, Y., Ma, Y., & Wang, Y. (2025) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2025.3547122 · Cited by 6
“Similarly, clustering algorithms based on superpixels can also increase the reliability of results [43], and using clustering results as pseudo-labels has been proved to be effective [44].”
Weakly Supervised Character Detection for License Plate Recognition
Zeni, L. F., & Jung, C. R. (2020) · SIBGRAPI Conference on Graphics, Patterns and Images (SIBGRAPI) · doi:10.1109/sibgrapi51738.2020.00037 · Cited by 7
“In [27], the authors improved the refinement process adding proposal clusters to select one or more supervision boxes during the training, whereas Zeni and Jung [13] proposed a refinement knowledge distillation module and an adaptive refinement supervision function to the OICR, called boosted OICR…”
BackgroundGao, H., Huang, S., Li, M., & Li, M. (2024) · IEEE Access · doi:10.1109/access.2024.3406968 · Cited by 7
“In recent years, significant progress has been made in pedestrian detection through the successful application of deep convolutional neural networks [1], [2], [3], [4], [5], [6], [7], [8], [9].”
BackgroundKanadath, A., Jothi, J. A. A., & Urolagin, S. (2025) · Computers in Biology and Medicine · doi:10.1016/j.compbiomed.2025.110680 · Cited by 5
Weakly Supervised Object Detection for Aerial Images With Instance-Aware Label Assignment
Xie, X., Yuan, X., & Cheng, G. (2025) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2025.3609322 · Cited by 5
“Extending this idea, Tang et al. [42] introduced a proposal clustering strategy, enabling the selection of multiple supervision boxes during training.”
InfluentialYang, P., Zhou, S., Wang, L., & Yang, G. (2023) · Multimedia Tools and Applications · doi:10.1007/s11042-023-17237-1 · Cited by 5
“Essentially, the refinement process in OICR [27] and PCL [28] is a kind of instance-level knowledge distillation.”
Liu, P., Pan, Z., Lei, B., & Hu, Y. (2024) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2024.3354999 · Cited by 5
“…revolves around solving the part domination in natural image set tasks, and meanwhile, other problems like instance ambiguity, memory consumption, augmentation consistency, etc., are studied intensively [2], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27].”
BackgroundA visual knowledge oriented approach for weakly supervised remote sensing object detection
Zhang, J., Ye, B., Zhang, Q., Gong, Y., Lu, J., & Zeng, D. (2024) · Neurocomputing · doi:10.1016/j.neucom.2024.128114 · Cited by 5
Minimizing Sample Redundancy for Label-Efficient Object Detection in Aerial Images
Zhang, R., Xu, C., Zhu, H., Xu, F., Yang, W. T., Zhang, H., & Xia, G.-S. (2025) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2025.3562395 · Cited by 11
“WSOD aims to achieve object detection with minimal supervision, such as image-level annotations [18], [19], [20], [21], [22], [23], [24], [25], [26], point annotations [27], [28], [29], [30], or scribbles [31], [32], [33], without relying on precise bounding box annotations.”
Single-shot weakly-supervised object detection guided by empirical saliency model
Zhao, D., Yuan, Z., Shi, Z., & Xie, F. (2021) · Neurocomputing · doi:10.1016/j.neucom.2021.03.047 · Cited by 5
Dynamic updating self-training for semi-weakly supervised object detection
Zhang, M., Liu, S., & Zeng, B. (2023) · Neurocomputing · doi:10.1016/j.neucom.2023.126375 · Cited by 6
Multi-Grained and Confidence-Aware Multiple Instance Network for Infrared Target Detection
Chen, W., Yang, H., Chang, S., Chen, Y., Wang, X., & Chen, Y. (2024) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2024.3422924 · Cited by 5
“Restrictions apply. to optical datasets with various target categories [19], [23], [28], [29], [30], [31], [32], [33], [34], [35].”
BackgroundMethodSelf-citationRegularized Instance Embedding for Deep Multi-Instance Learning
Lin, Y., & Zhang, H. (2019) · Applied Sciences · doi:10.3390/app10010064 · Cited by 5
“[4] used instance clustering to stabilize the process of weakly-supervised object detection.”
BackgroundMethodWeakly Supervised Temporal Action Localization by Multi-Stage Fusion Network
Shen, Z., Wang, F., & Dai, J. (2020) · IEEE Access · doi:10.1109/access.2020.2967627 · Cited by 5
Movable Object Detection in Remote Sensing Images via Dynamic Automatic Learning
Zhang, X., Luo, H., Zhong, S., Tang, L., Peng, J., & Jianping Fan, B. (2023) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2023.3270496 · Cited by 5
“To reduce the labeling cost, numerous methods [9], [10], [11], [12], [13], [14], [15], [16] Xiang Zhang, Hangzai Luo, Sheng Zhong, Jinye Peng, and Jianping Fan are with the School of Information and Technology, Northwest University, Xi’an 710127, China (e-mail: xiangz@nwu.edu.cn; hzluo@nwu.edu.cn;…”
MethodLai, Q., Vong, C., Shi, S.-Q., & Chen, C. L. P. (2024) · IEEE Transactions on Emerging Topics in Computational Intelligence · doi:10.1109/tetci.2024.3436853 · Cited by 5
“Following [4], [5], [16], the training set has 41 K images, the validation set has 5 K images, and the test set has about 20 K images.”
InfluentialHsu, C.-Y., & Li, W. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.04009 · Cited by 5
“This work shares the same goal as our proposed method; (4) C-WSL [12], which uses per-class count as the weak labels for WSOD; and three most recent state-of-the-art WSOD solutions: (5) C-MIDN [13], (6) C-MIL [28], and (7) MIL-OICR(PCL)+GAM+REG [31].”
InfluentialPersonalizing Pre-trained ModelsPreprint
Khan, M., Srivatsa, P., Rane, A., Chenniappa, S., Hazariwala, A., & Maes, P. (2021) · arXiv.org · doi:10.48550/arxiv.2106.01499 · Cited by 7
“, 2020a;c), PCL Resnet50 (Tang et al., 2018), SwAV Resnet50 (Caron et al.”
P2Object: Single Point Supervised Object Detection and Instance Segmentation
Chen, P., Yu, X., Han, X., Wang, K., Li, G., Xie, L., Han, Z., & Jiao, J. (2025) · International Journal of Computer Vision · doi:10.1007/s11263-025-02441-3 · Cited by 6
“InMIL-basedmethods(Bilen&Vedaldi,2016;Tang et al., 2017, 2020; Chen et al., 2020b; Wan et al., 2019), a bag is positively labeled if it contains at least one positive instance; otherwise, it is negative.”
InfluentialZhong, S., Song, P., Zhou, L., Gong, S., & Xie, G. (2023) · IEEE Transactions on Cognitive and Developmental Systems · doi:10.1109/tcds.2023.3312679 · Cited by 4
“For example, some studies have focused on generating more high-quality proposals [5], [15], [26], [31], while others have aimed to improve the way of pseudo-labeling of region instances [17], [20], [29], [41], [42], [44], [48].”
BackgroundMethodProposal Feature Learning Using Proposal Relations for Weakly Supervised Object Detection
Wang, Z., Zhang, W., & Zhang, M.-L. (2024) · IEEE International Conference on Multimedia and Expo (ICME) · doi:10.1109/icme57554.2024.10688185 · Cited by 7
“OICR [3] 37.9 62.1 PCL [18] 40.6 63.2 C-MIDN [19] 50.2 71.2 WSOD2 [20] 47.2 71.9 OIM [21] 45.3 67.1 SLV [22] 53.5 49.2 ICM-WSD [4] 52.1 70.9 CASD [23] 53.6 72.3 NDI-WSOD [5] 53.9 72.2 The results of our two methods on MS COCO is illusteated in Table III.”
InfluentialWS-SAM: Generalizing SAM to Weakly Supervised Object Detection With Category Label
Wang, H., Jia, T., Wang, Q., & Zuo, W. (2025) · IEEE Transactions on Image Processing · doi:10.1109/tip.2025.3581729 · Cited by 6
Li, S., Chang, B., Yang, B., Wu, H., Zhong, S., & Xu, F. (2023) · International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR) · doi:10.1145/3539618.3591661 · Cited by 4
“Based on this work, works such as [28, 29] add instance classifiers to make highly-overlapped object proposals share similar label information.”
BackgroundAthlete target detection method in dynamic scenario based on nonlinear filtering and YOLOv5
Dong, W., Pan, L., Zhang, Q., & Zhang, W. (2023) · Alexandria Engineering Journal · doi:10.1016/j.aej.2023.09.061 · Cited by 4
Hu, X., Zhu, M., Feng, Z., & Stanković, L. (2024) · IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing · doi:10.1109/jstars.2024.3459792 · Cited by 5
Zhang, M., & Zeng, B. (2020) · Computer Vision and Image Understanding · doi:10.1016/j.cviu.2020.102903 · Cited by 4
Performing Weakly Supervised Retail Instance Segmentation via Region Normalization
Bi, X., & Wang, L. (2021) · IEEE Access · doi:10.1109/access.2021.3077031 · Cited by 4
Online Progressive Instance-Balanced Sampling for Weakly Supervised Vibration Damper Detection
Chen, M., Tian, Y., Li, Z., Li, E., & Liang, Z. (2023) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2023.3273655 · Cited by 5
“Based on this algorithm, some methods [15], [20], [21], [22] further get much better results by complicating the selection of positive instances or adding extra branches of classification, which lead to high time and space complexity.”
BackgroundMethodA Robust Context-Aware Proposal Refinement Method for Weakly Supervised Object Detection
Awan, M., & Shin, J. (2020) · IEEE Access · doi:10.1109/access.2020.3035606 · Cited by 4
“The approaches which integrate the CNN with MIL for WSOD have shown performance improvement by CNN as feature extractor compared to conventional hand-crafted features [13].”
BackgroundMethodInfluentialEnabling Near-Zero Cost Object Detection in Remote Sensing Imagery via Progressive Self-Training
Zhang, X., Jiang, X., Hu, Q., Luo, H., Zhong, S., Tang, L., Peng, J., & Jianping Fan, B. (2024) · IEEE Transactions on Geoscience and Remote Sensing · doi:10.1109/tgrs.2024.3415002 · Cited by 4
Wu, Z., Zheng, S., Xu, Y., Wang, L., Wei, Z., & Hua, G. (2025) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2025.3646464 · Cited by 3
“Specifically, multi-stage proposal-based [15], [16], [17], [18], [19], [20], [34], [36], [39], [41], [50], [51] and end-to-end CAM-based [26], [27], [28]methodsareallcomparedwithourBSANet.”
SAPNet++: Evolving Point-Prompted Instance Segmentation With Semantic and Spatial Awareness
Wei, Z., Han, X., Yu, X., Yang, X., Li, G., Han, Z., & Jiao, J. (2026) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2026.3667694 · Cited by 3
“Inspired by PCL [68], we introduce a two-stage refinement strategy.”
Dynamic supervisor for cross-dataset object detection
Chen, Z., Fu, Z., Huang, J., Tao, M., Li, S., Jiang, R., Tian, X., Chen, Y., & Hua, X. (2021) · Neurocomputing · doi:10.1016/j.neucom.2021.09.076 · Cited by 3
“These incomplete annotation sets further transform fully supervised learning into weakly [9, 10, 11, 12] or semi-supervised learning [13, 14, 15].”
Complementary Cues from Audio Help Combat Noise in Weakly-Supervised Object Detection
Gungor, C., & Kovashka, A. (2023) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv56688.2023.00222 · Cited by 3
“We compare our detection performance with state-of-art detectors Afouras [1] and PCL [35] in Table 2.”
BackgroundMethodInfluentialA Pedestrian Detection Network Based on an Attention Mechanism and Pose Information
Jiang, Z., Huang, S., & Li, M. (2024) · Applied Sciences · doi:10.3390/app14188214 · Cited by 4
Image-caption difficulty for efficient weakly-supervised object detection from in-the-wild data
Nebbia, G., & Kovashka, A. (2024) · IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) · doi:10.1109/cvprw63382.2024.00266 · Cited by 4
“One of the first successful approaches for WSOD is Weakly Supervised Deep Detection Networks (WSDDN) [3], from which numerous models originated [24, 32, 33, 38].”
Liu, P., Huang, B., Jin, T., & Long, H. (2025) · Remote Sensing · doi:10.3390/rs17071155 · Cited by 3
Cui, Z., Lu, C., Liu, Y., & Yin, Z. (2025) · Pattern Recognition · doi:10.1016/j.patcog.2025.112707 · Cited by 3
Point in: Counting Trees with Weakly Supervised Segmentation Network
Tong, P., Zhang, X., Han, P., & Bu, S. (2021) · Proceedings - International Conference on Pattern Recognition/Proceedings/International Conference on Pattern Recognition · doi:10.1109/icpr48806.2021.9412397 · Cited by 3
“In addition, we also compare with some weakly supervised methods based on image level annotations, like WSDDN [14], PCL [26], and C-MIL [15].”
InfluentialHow to Save your Annotation Cost for Panoptic Segmentation?
Du, X., Jiang, C., Xu, H., Zhang, G., & Li, Z. (2021) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v35i2.16216 · Cited by 3
“For object detection, such approaches train a detector using image labels by Multiple Instance Learning (MIL) (Zhang et al. 2018a; Jie et al. 2017; Tang et al. 2017, 2020).”
Method
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.