Multilabel learning based adaptive graph convolutional network for human parsing

作者：

Highlights：

• The fixed graph modeling may not be an optimal graph for the diversity of the samples.

• A multilabel learning based adaptive graph generation module generate the adaptive graph.

• A semantic parts based attention module fuse fixed graph and adaptive graph to obtain a more comprehensive graph.

• The label consistency loss explicitly constraint the consistency between the predicted multilabel and the predicted human parsing results.

摘要

•The fixed graph modeling may not be an optimal graph for the diversity of the samples.•A multilabel learning based adaptive graph generation module generate the adaptive graph.•A semantic parts based attention module fuse fixed graph and adaptive graph to obtain a more comprehensive graph.•The label consistency loss explicitly constraint the consistency between the predicted multilabel and the predicted human parsing results.

论文关键词：Human parsing,Multilabel learning based adaptive graph convolutional network,Adaptive graph

论文评审过程：Received 25 July 2021, Revised 22 January 2022, Accepted 16 February 2022, Available online 18 February 2022, Version of Record 24 February 2022.

论文官网地址：https://doi.org/10.1016/j.patcog.2022.108593