论文笔记: Fully Convolutional Networks for Semantic Segmentation

想法

Our key insight is to build “fully convolutional” networks that take arbitrary size and produce correspondingly-sized output with efficient inference and learning.

Convnets are built on translation invariance. Their basic components (convolution, pooling, and activation functions) operate on local input regions, and depend only on relative spatial coordinates.

变成全卷积的效果图

可以看出有原来的 softmax 之类的响应，变成了heatmap

具体操作

得到1000个feature map 之后进行上采样成和原图像相同大小的feature map，这样就可以进行用原图直接训练。上采样用反卷积。为了保留低层feature map的特征，作者用了下面：



虽然挺好，但是导师一直有个思想—-这样的操作违背了深度学习的初衷，就要高层提取什么低层信息。