Vehicle joint make and model recognition with multiscale attention windows

作者：

Highlights：

• VMMR can be addressed by formulating a joint loss function.

• Predicting attention window scale improves VMMR Performance.

• Multi-scale visual representations can increase VMMR accuracy.

• Multi-scale patch training enables generating multi-scale attention windows.

摘要

•VMMR can be addressed by formulating a joint loss function.•Predicting attention window scale improves VMMR Performance.•Multi-scale visual representations can increase VMMR accuracy.•Multi-scale patch training enables generating multi-scale attention windows.

论文关键词：Vehicle classification,Convolutional neural network,Attention windows,Residual network

论文评审过程：Received 26 July 2018, Revised 9 November 2018, Accepted 12 December 2018, Available online 22 December 2018, Version of Record 29 December 2018.

论文官网地址：https://doi.org/10.1016/j.image.2018.12.009