Caffe Introduction to basic knowledge

A lot of good pages:

1, http://alanse7en.github.io/caffedai-ma-jie-xi-1/mainly introduce basic caffe knowledge

The ' I ' in the Interace interface API

The Caffe code structure consists of four large classes: Solver: The optimization function encapsulation, which has a protected member shared_ptr<net> net_; This member is a pointer to Net type only, Solver uses this pointer to interact with the network net and to complete the optimization of the model. Different sub-classes have implemented different optimization methods: Sgdsolver nesterovsolver, Adagradsolver, Rmspropsolver, Adadeltasolver nuclear adamsolver.

Net: Complete the encapsulation of the whole network, which has a member of:vector<shared_ptr<layer>> Layers_; This vector contains a smart pointer for each layer in the entire network, net by invoking these layers ' respective forward () and backward () interfaces to achieve the overall network

Layer: Similar to layer this class derives a number of subclasses that implement the layer of data, such as the read kernel convolution,pooling,innerproduct.

Blob: Encapsulation of Caffe data, in the entire network calculation, whether the data or network parameters of the kernel gradient is the object of this class, are num*channel*width*height form of data

2, http://www.voidcn.com/blog/iamzhangzhuping/list-17.html

Program Park is very good blog site, and the blog park is not the same ha

3.

Caffe Introduction to basic knowledge