摘要
图像语义分割是计算机视觉领域重要识别任务,其目标是估计图像中的像素级目标类标签。最近,深度卷积神经网络(Deep Convolutional Neural Networks,DCNNs)已经成为解决图像语义分割的主流方法。然而,学习DCNNs需要大量的已标注训练数据(Ground Truth,GT),而现有数据集中的GT在数量和多样性方面因标注成本巨大而受到诸多限制。弱监督方法则考虑利用图像级标签和物体框之类的弱标注信息解决图像语义分割中的标注问题。相比于全监督的像素级图像标注,图像分类的GT(图像级标签)和目标检测的GT(物体框)更容易获得,因而可以直接借用为弱标注信息训练分类模型。弱监督语义分割的主要挑战在于标注信息的不完整性,即缺失了物体精确的边界信息。文中对基于DCNNs的弱监督语义分割方法进行了全面的阐述,描述了如何克服这些限制并讨论了提高其性能的可能研究方向。
Image semantic segmentation is an important visual recognition task and its goal is to estimate pixel-level object class labels on images. This problem has been recently handled by deep convolutional neural networks（DCNNs）. However,learning DCNNs demand a large number of annotated training data while segmentation annotations in existing data sets are limited in terms of both quantity and diversity due to the heavy annotation cost. Weakly supervised approaches tackle this issue by leveraging weak annotations,such as image-level labels and bounding boxes,which are either readily available in existing largescale data sets for image classification and object detection. The main challenge in weakly supervised semantic segmentation is the incomplete annotations,missing accurate object boundary information required to learn segmentation. A comprehensive overview of weakly supervised approaches for semantic segmenta-tion is provided. The approaches for overcome the limitations are decribed and research directions of improving performances are pointed out.
作者
郑宝玉
王雨
吴锦雯
周全
ZHENG Baoyu;WANG Yu;WU Jinwen;ZHOU Quan(College of Telecommunications ＆ Information Engineering,Nanjing University of Posts and Telecommunications,Nanjing 210003,China;Engineering Research Center for Communication and Network,Nanjing University of Posts and Telecommunications,Nanjing 210003,China;College of Information Engineering,China University of Geoseienees（Wuhan）,Wuhan 430074,China)
出处
《南京邮电大学学报（自然科学版）》
北大核心
2018年第5期1-12,共12页
Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition
基金
国家自然科学基金(61876093
61881240048
61671253
61701252
61762021)
江苏省自然科学基金(BK20181393
BK20160908)资助项目
关键词
语义分割
深度卷积神经网络
弱监督语义分割
图像标注
semantic segmentation
deep convolutional neural networks（DCNNs）
weakly supervised se-mantic segmentation
image annotation
