- CNN - # 2.Padding (Andrew NG）
- 机器学习笔记 01 Logistic Regression
- stanford公开课Machine Learning视频教程
- Machine Learning-bias & vari
- Deep Learning | 2 Improving Deep
- Machine Learning Series Notes -
- 2018-06-14 Machine Learning by A
- Deep Learning-complex structure
- Machine Learning @ Coursera
- Machine Learning by Andrew Ng -
2.1 Padding
original image * filter 的缺点：
- shrinking output 导致生成的图缩小
- throw away info from edge 边缘的像素点的信息只被用到了相对较少的次数，甚至有的只有1次。相比起居中的像素点，这丢掉了一些边缘像素点的信息
So, 为什么需要padding?
===> padding 将边缘的框框包裹1层或多层，使得原本edge pixels不再那么边缘，保住了info.比如以前的6x6的original image,做了padding之后，再乘以filter,生成了和original image一样size的matrix.
卷积有2种:
- valid
- same
(Main contents are extracted from Andrew NG's deep learning course:https://mooc.study.163.com/learn/2001281004?tid=2001392030#/learn/content?type=detail&id=2001729325&cid=2001724505)
网友评论