SegNet: a deep convolutional encoder-decoder architecture for robust semantic pixel-wise labelling

框架：SegNet， a deep convolutional encoder-decoder architecture

框架组成部分：A stack of encoders

                       + a corresponding decoder stack

                       + then feed into a softmax classification layer

应用：semantic pixel wise image labeling

三个亮点：1、only require forward evaluation of a fully learnt function to obtain smooth label predictions

                     （仅需要一个前向估计函数就可以获得平滑标签预测）

                2、with increasing depth (of neural network), a larger context is considered for pixel labeling which improves accuracy. 

                     (网络深度增加，像素点context信息能够考虑到)

                3、be easy to visualizes the effect of feature activations in the pixel labels space at any depth.

                     (在网络的任何一层，都可以可视化特征)

Framework