https://www.zhihu.com/question/66283266/answer/240344515
第一句
We adapted the VGG-16 network (Simonyan & Zisserman, 2015) for dense prediction and removed the last two pooling and striding layers.
第二句
Specifically, each of these pooling and striding layers was removed and convolutions in all subsequent layers were dilated by a factor of 2 for each pooling layer that was ablated.
第三句
Most significantly, we remove the last two pooling and striding layers entirely, whereas Long et al. kept them and Chen et al. replaced striding by dilation but kept the pooling layers.
作者把pooling和striding两个单词是一起用，也即是卷积网络中池化和子采样层是分别实现的，当然现在一般理解为一起，因为pooling中一般stride=2。那么pooling中的stride=1，结合padding的话，完全可以做到保持feature maps大小不变。其实pooling完全可以理解为conv。
作者：小锋子Shawn
链接：https://www.zhihu.com/question/66283266/answer/240344515
来源：知乎
著作权归作者所有。商业转载请联系作者获得授权，非商业转载请注明出处。