计算机工程与应用 ›› 2021, Vol. 57 ›› Issue (4): 43-54.DOI: 10.3778/j.issn.1002-8331.2010-0127

• 热点与综述 • 上一篇    下一篇

全监督学习的图像语义分割方法研究进展

袁铭阳,黄宏博,周长胜   

  1. 1.北京信息科技大学 计算机学院,北京 100101
    2.北京信息科技大学 计算智能研究所,北京 100192
  • 出版日期:2021-02-15 发布日期:2021-02-06

Research Progress of Image Semantic Segmentation Based on Fully Supervised Learning

YUAN Mingyang, HUANG Hongbo, ZHOU Changsheng   

  1. 1.School of Computer Science, Beijing Information Science and Technology University, Beijing 100101, China
    2.Institute of Computational Intelligence, Beijing Information Science and Technology University, Beijing 100192, China
  • Online:2021-02-15 Published:2021-02-06

摘要:

近年来,随着深度学习进入计算机视觉领域,各种深度学习图像语义分割方法相继出现,其中全监督学习方法的分割效果显著超过弱监督学习方法。将全监督学习的图像语义分割方法分为五类,并对各类中最具有代表性的方法进行详细分析,重点阐述各种方法核心部分的实现过程。对语义分割领域中的主流数据集进行归纳总结,介绍了性能算法指标,并在主流数据集上对各种代表性方法的效果进行对比,最后对语义分割的未来进行展望。

关键词: 计算机视觉, 图像语义分割, 深度学习, 语义分割方法, 全监督学习

Abstract:

In recent years, as deep learning enters the field of computer vision, various deep learning image semantic segmentation methods that are superior to traditional methods have appeared one after another, including fully supervised learning, the segmentation effect of the method significantly exceeds that of weakly supervised learning method, the image semantic segmentation methods of fully supervised learning are divided into five categories, and the most representative methods of the various types are analyzed in detail, and the implementation processes of the core components of each method are emphasized. After that, this paper summarizes the mainstream data sets in the field of semantic segmentation, introduces the performance algorithm indicators, and compares the effects of various representative methods on the mainstream data sets, and finally looks forward to the future of semantic segmentation.

Key words: computer vision, image semantic segmentation, deep learning, semantic segmentation method, fully supervised learning