Unbiased Subclass Regularization for Semi-Supervised Semantic Segmentation

被引:21
|
作者
Guan, Dayan [1 ]
Huang, Jiaxing [1 ]
Xiao, Aoran [1 ]
Lu, Shijian [1 ]
机构
[1] Nanyang Technol Univ, Singtel Cognit & Artificial Intelligence Lab Ente, Singapore, Singapore
关键词
D O I
10.1109/CVPR52688.2022.00973
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Semi-supervised semantic segmentation learns from small amounts of labelled images and large amounts of unlabelled images, which has witnessed impressive progress with the recent advance of deep neural networks. However, it often suffers from severe class-bias problem while exploring the unlabelled images, largely due to the clear pixel-wise class imbalance in the labelled images. This paper presents an unbiased subclass regularization network (USRN) that alleviates the class imbalance issue by learning class-unbiased segmentation from balanced subclass distributions. We build the balanced subclass distributions by clustering pixels of each original class into multiple subclasses of similar sizes, which provide class-balanced pseudo supervision to regularize the class-biased segmentation. In addition, we design an entropy-based gate mechanism to coordinate learning between the original classes and the clustered subclasses which facilitates subclass regularization effectively by suppressing unconfident subclass predictions. Extensive experiments over multiple public benchmarks show that USRN achieves superior performance as compared with the state-of-the-art.
引用
收藏
页码:9958 / 9968
页数:11
相关论文
共 50 条
  • [21] Semi-Supervised Semantic Segmentation with Cross Pseudo Supervision
    Chen, Xiaokang
    Yuan, Yuhui
    Zeng, Gang
    Wang, Jingdong
    2021 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION, CVPR 2021, 2021, : 2613 - 2622
  • [22] Semi-supervised Learning for Segmentation Under Semantic Constraint
    Ganaye, Pierre-Antoine
    Sdika, Michael
    Benoit-Cattin, Hugues
    MEDICAL IMAGE COMPUTING AND COMPUTER ASSISTED INTERVENTION, PT III, 2018, 11072 : 595 - 602
  • [23] Semi-supervised Semantic Segmentation with Complementary Reconfirmation Mechanism
    Xiao, Yifan
    Dong, Jing
    Zhang, Qiang
    Yi, Pengfei
    Liu, Rui
    Wei, Xiaopeng
    ADVANCES IN COMPUTATIONAL INTELLIGENCE SYSTEMS, UKCI 2023, 2024, 1453 : 182 - 194
  • [24] Enhanced Soft Label for Semi-Supervised Semantic Segmentation
    Ma, Jie
    Wang, Chuan
    Liu, Yang
    Lin, Liang
    Li, Guanbin
    2023 IEEE/CVF INTERNATIONAL CONFERENCE ON COMPUTER VISION, ICCV, 2023, : 1185 - 1195
  • [25] Semi-supervised Semantic Segmentation with Error Localization Network
    Kwon, Donghyeon
    Kwak, Suha
    2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), 2022, : 9947 - 9957
  • [26] An efficient and scalable semi-supervised framework for semantic segmentation
    Huazheng Hao
    Hui Xiao
    Junjie Xiong
    Li Dong
    Diqun Yan
    Dongtai Liang
    Jiayan Zhuang
    Chengbin Peng
    Neural Computing and Applications, 2025, 37 (7) : 5481 - 5497
  • [27] Fuzzy Positive Learning for Semi-supervised Semantic Segmentation
    Qiao, Pengchong
    Wei, Zhidan
    Wang, Yu
    Wang, Zhennan
    Song, Guoli
    Xu, Fan
    Ji, Xiangyang
    Liu, Chang
    Chen, Jie
    2023 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), 2023, : 15465 - 15474
  • [28] Colour Augmentation for Improved Semi-supervised Semantic Segmentation
    French, Geoff
    Mackiewicz, Michal
    PROCEEDINGS OF THE 17TH INTERNATIONAL JOINT CONFERENCE ON COMPUTER VISION, IMAGING AND COMPUTER GRAPHICS THEORY AND APPLICATIONS (VISAPP), VOL 4, 2022, : 356 - 363
  • [29] Semi-supervised Semantic Segmentation with Mutual Knowledge Distillation
    Yuan, Jianlong
    Ge, Jinchao
    Wang, Zhibin
    Liu, Yifan
    PROCEEDINGS OF THE 31ST ACM INTERNATIONAL CONFERENCE ON MULTIMEDIA, MM 2023, 2023, : 5436 - 5444
  • [30] Catastrophic Forgetting Problem in Semi-Supervised Semantic Segmentation
    Zhou, Yan
    Jiao, Ruyi
    Wang, Dongli
    Mu, Jinzhen
    Li, Jianxun
    IEEE ACCESS, 2022, 10 (48855-48864) : 48855 - 48864