解读Depth Map Prediction from a Single Image using a Multi-Scale Deep Network （2）
把CNN的基本知识补全后，接着向下看 Section 3 Approach
卷积网络分为两个部分：
a. A coarse-scale network predicts the depth of the scene at a global level
b.It is then refined within local regions by a fine-scale network
如下图所示，
先来看 Global Coarse-Scale Network
主要作用：predict the overall depth map structure using a global view of the scene
卷积层叙述：
upper layers(fully connected): contain the entire image in their field of view
lower and middle layers : contain information from different parts of the image
卷积层设计的特点：
a. be able to integrate a global understanding of the full scene to predict the depth
b. make effective use of cues such as vanishing point, object location
再来看 Local Fine-Scale Network
主要作用：to edit the coarse prediction it receives to align with local details such as objects and wall edges
卷积层特点：
a. consists of convolutional layers only, along with one pooling stage for the first layer edge features
b. Subsequent layers maintain this size using zero-padded convolutions
c. All hidden units use rectified linear activations
CNN网络训练过程简述：
a. first train the coarse network against the ground-truth targets
b. then train the fine-scale network keeping the coarse-scale output fixed
(when training the fine network, we do not backpropagate through the coarse one)
CNN卷积神经网络的训练类似于传统BP神经网络的训练，即
1，需要定义网络学习参数以及误差函数
2，推导出参数权值更替的表达式
这一方面的知识需要补充！
下一次主要学习CNN卷积神经网络的反向传播过程！