计算机科学 ›› 2023, Vol. 50 ›› Issue (10): 1-6.doi: 10.11896/jsjkx.230600035
宋法兴, 苗夺谦, 张红云
SONG Faxing, MIAO Duoqian, ZHANG Hongyun
摘要: 深度学习对大规模数据的需求以及目标检测标注任务的复杂性促进了半监督目标检测任务的发展。近年来,半监督目标检测已经取得了很多优秀的成果。然而,伪标签中的不确定性依然是半监督目标检测研究中难以避免的问题,优越的半监督方法要求选取合适的过滤阈值来权衡伪标签的噪声信息比例和召回率,以最大程度保留准确有效的伪标签。为了解决此问题,在半监督检测的框架中引入了序贯三支决策算法,将模型输出的伪标签根据不同的筛选阈值划分为干净的前景标签、有噪声的前景标签,以及干净的背景标签,并对其采取不同的处理策略。对有噪声的前景标签采用负类学习损失来学习这些存在噪声的标签,避免学习到其中的噪声信息。实验结果表明了所提算法的性能优势,针对COCO数据集,在有监督数据占比只有10%的情况下,该方法实现了35.2%的检测精度,相比仅依靠有监督训练性能提升了11.34%。
中图分类号:
[1]XU Y,SHANG L,YE J,et al.Dash:Semi-supervised learningwith dynamic thresholding[C]//International Conference on Machine Learning(ICML).Cambridge MA:JMLR,2021:11525-11536. [2]YUE X D,CHEN Y F,MIAO D Q,et al.Fuzzy NeighborhoodCovering for Three-way Classification[J].Information Sciences,2020,507:795-808. [3]WEI X S,XU H Y,ZHANG F,et al.An Embarrassingly Simple Approach to Semi-Supervised Few-Shot Learning[J].Advances in Neural Information Processing Systems,2022,35:14489-14500. [4]LIN T Y,MAIRE M,BELONGIE S,et al.Microsoft COCO:Common Objects in Context[C]//European Conference on Computer Vision(ECCV).Cham:Springer,2014:740-755. [5]LIU W,ANGUELOV D,ERHAN D,et al.Ssd:Single shotmultibox detector[C]//European Conference on Computer Vision(ECCV).Cham:Springer,2016:21-37. [6]LIN T Y,GOYAL P,GIRSHICK R,et al.Focal Loss for Dense Object Detection[C]// Conference on Computer Vision(ICCV).Cham:Springer,2017:2980-2988. [7]GIRSHICK R,DONAHUE J,DARRELL T,et al.Rich feature hierarchies for accurate object detection and semantic segmentation[C]// Computer Vision and Pattern Recognition(CVPR).NJ:IEEE,2014:1714-1722. [8]GIRSHICK R.Fast R-CNN[C]// International Conference on Computer Vision(ICCV).Cham:Springer,2015:1440-1448. [9]REN S,HE K M,GIRSHICK R,et al.Faster R-CNN:Towards Real-Time Object Detection with Region Proposal Networks[J].IEEE Transactions on Pattern Analysis & Machine Intelligence,2017,39(6):1137-1149. [10]DUAN K W,BAI S,XIE L X,et al.CenterNet:Keypoint Triplets for Object Detection[C]// IEEE/CVF International Confe-rence on Computer Vision(ICCV).Cham:Springer,2019:1-16. [11]TIAN Z,SHEN C H,CHEN H,et al.FCOS:Fully Convolu-tional One-Stage Object Detection[C]// International Confe-rence on Computer Vision(ICCV).NJ:IEEE,2019:9627-9636. [12]PHAM H,DAI Z,XIE Q,et al.Meta pseudo labels[C]//Proceedings of the IEEE/CVF conference on Computer Vision and Pattern Recognition(CVPR).NJ:IEEE,2021:11557-11568. [13]XIE Q,LUONG M T,HOVY E,et al.Self-training with noisy student improves imagenet classification[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).NJ:IEEE,2020:10687-10698. [14]ZOPH B,GHIASI G,LIN T Y,et al.Rethinking pre-trainingand self-training[J].Advances in Neural Information Processing Systems,2020,33:3833-3845. [15]DEVRIES T,TAYLOR G W.Improved regularization of convolutional neural networks with cutout[J].arXiv:1708.04552,2017. [16]ZHANG H,CISSE M,DAUOHIN Y N,et al.mixup:Beyond empirical risk minimization[J].arXiv:1710.09412,2017. [17]YUN S,HAN D,OH S J,et al.Cutmix:Regularization strategy to train strong classifiers with localizable features[C]//Procee-dings of the IEEE/CVF International Conference on Computer Vision(ICCV).NJ:IEEE,2019:6023-6032. [18]SOHN K,ZHANG Z,LI C L,et al.A simple semi-supervisedlearning framework for object detection[J].arXiv:2005.04757,2020. [19]LI H,WU Z,SHRIVASTAVA A,et al.Rethinking pseudo labels for semi-supervised object detection[C]//Proceedings of the AAAI Conference on Artificial Intelligence(AAAI).CA:AAAI,2022:1314-1322. [20]KIM J M,JANG J Y,SEO S,et al.MUM:Mix Image Tiles and UnMix Feature Tiles for Semi-Supervised Object Detection[C]//2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).NJ:IEEE,2022:14492-14501. [21]CHEN B,LI P,CHEN X,et al.Dense learning based semi-supervised object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).NJ:IEEE,2022:4815-4824. [22]CHEN C,DEBATTISTA K,HAN J.Semi-supervised object detection via virtual category learning[J].arXiv:2207.03433,2022. [23]XU M,ZHGANG Z,HU H,et al.End-to-end semi-supervisedobject detection with soft teacher[C]//Proceedings of the IEEE/CVF International Conference on Computer Vision(ICCV).NJ:IEEE,2021:3060-3069. [24]EVERINGHAM M,VAN GOOL L,WILLIAMS C K I,et al.The pascal visual object classes(voc) challenge[J].International Journal of Computer Vision,2009,88:303-308. [25]CHEN K,WANG J,PANG J,et al.MMDetection:Open mmlab detection toolbox and benchmark[J].arXiv:1906.07155,2019. [26]HE K,ZHANG X,REN S,et al.Deep residual learning forimage recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.2016:770-778. [27]JEONG J,LEE S,KIM J,et al.Consistency-based semi-super-vised learning for object detection[C]//ICCV 2019.2019. [28]ZHOU Q,YU C,WANG Z,et al.Instant-teaching:An end-to-end semi-supervised object detection framework[C]//Procee-dings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).NJ:IEEE,2021:4081-4090. [29]YANG Q,WEI X,WANG B,et al.Interactive self-training with mean teachers forsemi-supervised object detection[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).NJ:IEEE,2021:5941-5950. [30]TANG Y,CHEN W,LUO Y,et al.Humble teachers teach better students for semi-supervised object detection[C]//Procee-dings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).NJ:IEEE,2021:3132-3141. [31]LIU Y C,MA C Y,HE Z,et al.Unbiased teacher for semi-supervised object detection[J].arXiv:2102.09480,2021. |
|