Incorporating part-whole hierarchies into fully convolutional network for scene parsing
作者：
Highlights：
• In this paper, a new layer is proposed which is called Caps-Score layer.
• The general Hough transform (GHT) is utilized to propose a novel capsule concept.
• To accumulate the vote’s strengths for capsule, tensor normal distribution is utilized.
• The covariance matrix is defined as Kronecker product of two sub-covariance matrices.
• The proposed layer handles translation and rotation discrepancies.
摘要
•In this paper, a new layer is proposed which is called Caps-Score layer.•The general Hough transform (GHT) is utilized to propose a novel capsule concept.•To accumulate the vote’s strengths for capsule, tensor normal distribution is utilized.•The covariance matrix is defined as Kronecker product of two sub-covariance matrices.•The proposed layer handles translation and rotation discrepancies.
论文关键词：Scene parsing,Caps-score layer,Fully convolutional network,The spatial hierarchy between features
论文评审过程：Received 13 July 2019, Revised 8 May 2020, Accepted 12 June 2020, Available online 21 June 2020, Version of Record 6 July 2020.
论文官网地址：https://doi.org/10.1016/j.eswa.2020.113662
