论文题目:End-to-end representation learning for Correlation Filter based tracking, CVPR2017
论文主页:http://www.robots.ox.ac.uk/~luca/cfnet.html
源码链接:https://github.com/bertinetto/cfnet
相关推导:https://blog.csdn.net/discoverer100/article/details/80030294#commentBox (非常详细)
1. 论文摘要、贡献
图一:CFNet整体架构。非对称的孪生网络,非对称部分是由于correlation filter部分引起的,作者是想把cf引入到神经网络中,从而能够实现end-to-end优化。training images和testing images首先都经过相同的卷积特征提取网络提取特征,其中training images提取出来的卷积特征进而通过correlation filter得到线性模板,