什么是卷积

目录

* Feature maps
<https://www.cnblogs.com/nickchen121/p/10923353.html#feature-maps>
* Why not Linear
<https://www.cnblogs.com/nickchen121/p/10923353.html#why-not-linear>
* 335k or 1.3MB
<https://www.cnblogs.com/nickchen121/p/10923353.html#k-or-1.3mb>
* em... <https://www.cnblogs.com/nickchen121/p/10923353.html#em...>
* Receptive Field
<https://www.cnblogs.com/nickchen121/p/10923353.html#receptive-field>
* Fully connnected
<https://www.cnblogs.com/nickchen121/p/10923353.html#fully-connnected>
* Partial connected
<https://www.cnblogs.com/nickchen121/p/10923353.html#partial-connected>
* Locally connected
<https://www.cnblogs.com/nickchen121/p/10923353.html#locally-connected>
* Rethink Linear layer
<https://www.cnblogs.com/nickchen121/p/10923353.html#rethink-linear-layer>
* Fully VS Lovally
<https://www.cnblogs.com/nickchen121/p/10923353.html#fully-vs-lovally>
* Weight sharing
<https://www.cnblogs.com/nickchen121/p/10923353.html#weight-sharing>
* Why call Convolution？
<https://www.cnblogs.com/nickchen121/p/10923353.html#why-call-convolution>
* 2D Convolution
<https://www.cnblogs.com/nickchen121/p/10923353.html#d-convolution>
* Convolution in Computer Vision
<https://www.cnblogs.com/nickchen121/p/10923353.html#convolution-in-computer-vision>
* CNN on feature maps
<https://www.cnblogs.com/nickchen121/p/10923353.html#cnn-on-feature-maps>
Feature maps

* 单通道


* rgb三通道


* rgb三通道合成


* 数字2的卷积成像图


Why not Linear

* 4 Layers: [784, 256, 256, 256, 10]


335k or 1.3MB



em...

* 486 PC + AT&T DSP32C
* 256KB
* 66Mhz
*
Batch X

*
Gradient Cache

*
etc.



Receptive Field



Fully connnected



Partial connected



Locally connected



Rethink Linear layer



Fully VS Lovally



Weight sharing



* 三阶张量的卷积


* 6 Layers
* ~60k parameters
* 4 layers, 335k


Why call Convolution？



2D Convolution

\[ y(t) = x(t)*h(t) = \int_{-\infty}^{\infty}x(\tau)h(t-\tau)d\tau \]



Convolution in Computer Vision



* 模糊化


* 边缘检测


CNN on feature maps

• « 上一篇：Python与用户交互
• » 下一篇：关于VM安装windows出现“Windows找不到Microsoft软件安装条款”解决办法