Ventral & Dorsal Stream Theory based Zero-Shot Action Recognition
作者：
Highlights：
• A general joint-conditional-probability framework is proposed to explain the inference mechanism of ZSAR methods. As far as we know, we are the first to propose the probabilistic model for the inference mechanism of ZSAR methods.
• A novel nonlinear similarity metric learning mechanism is proposed to establish the nonlinear mapping relationship between the visual space and the semantic space.
• A Ventral & Dorsal Stream Theory based Zero-Shot Action Recognition approach (VD-ZSAR) is proposed and achieved favorable performance on three benchmark datasets. The Ventral & Dorsal Stream Theory is introduced to extract irredundant visual feature, which can relieve relation ambiguity caused by redundant visual feature.
摘要
•A general joint-conditional-probability framework is proposed to explain the inference mechanism of ZSAR methods. As far as we know, we are the first to propose the probabilistic model for the inference mechanism of ZSAR methods.•A novel nonlinear similarity metric learning mechanism is proposed to establish the nonlinear mapping relationship between the visual space and the semantic space.•A Ventral & Dorsal Stream Theory based Zero-Shot Action Recognition approach (VD-ZSAR) is proposed and achieved favorable performance on three benchmark datasets. The Ventral & Dorsal Stream Theory is introduced to extract irredundant visual feature, which can relieve relation ambiguity caused by redundant visual feature.
论文关键词：VD-ZSAR,Ventral & Dorsal Stream Theory,Nonlinear similarity metric learning mechanism
论文评审过程：Received 30 June 2020, Revised 21 January 2021, Accepted 18 March 2021, Available online 28 March 2021, Version of Record 3 April 2021.
论文官网地址：https://doi.org/10.1016/j.patcog.2021.107953
