siamese network（暹罗网络）

缘起于寻找caffe如何输入多通道图片（两张四通道图片），希望通过寻找制作lmdb时，遇到的datum，来得到如何设置lmdb的通道。结果发现了siamese网络。

摘抄自caffe github的issue697

Siamese nets are supervised models for metric learning [1].

[1] S. Chopra, R. Hadsell, and Y. LeCun. Learning a similarity metric discriminatively, with application to face verification. In Computer Vision and Pattern Recognition, 2005. CVPR 2005. IEEE Computer Society Conference on, volume 1, pages 539–546. IEEE, 2005. http://yann.lecun.com/exdb/publis/pdf/chopra-05.pdf

 

Speaking of metric learning, I remember that @norouzi had proposed and open sourced a method that learned a Hamming distance metric to distinguish similar and dissimilar images [2].

[2] Mohammad Norouzi, David J. Fleet, Ruslan Salakhutdinov, Hamming Distance Metric Learning, Neural Information Processing Systems (NIPS), 2012.

 

 

以及一些论文需要看看：

2015CVPR：

[1]MatchNet:Unifying Feature and Metric Learning for Patch-Based Matching

[2]Learning to Compare Image Patches via Convolutional Neural Network

[3]Image Patch Matching Using Convolutional Descriptors with Euclidean Distance

大体上说，这些论文的思路都是利用一对CNN来提取一对图像特征，然后通过欧氏距离（经典如Siamese网络）或通过全连接网络（MatchNet）来实现特征的对比，最后通过交叉熵函数来完成优化。