Multi-view feature fusion for person re-identification

作者：

Highlights：

• The complementary-view features are defined to mitigate view bias.

• Multi-view Message Passing (MVMP) scheme generates multi-view features in the test stage.

• Multi-view Feature Fusion Network (MFFN) increases sensitivity to potential view-specific cues during training.

• Both MVMP and MFFN are parameter-free and can be applied to any Re-ID method readily without extra supervision.

摘要

•The complementary-view features are defined to mitigate view bias.•Multi-view Message Passing (MVMP) scheme generates multi-view features in the test stage.•Multi-view Feature Fusion Network (MFFN) increases sensitivity to potential view-specific cues during training.•Both MVMP and MFFN are parameter-free and can be applied to any Re-ID method readily without extra supervision.

论文关键词：Person re-identification,Convolutional Neural Networks,Message passing

论文评审过程：Received 24 January 2021, Revised 25 June 2021, Accepted 23 July 2021, Available online 10 August 2021, Version of Record 19 August 2021.

论文官网地址：https://doi.org/10.1016/j.knosys.2021.107344