Convolution and transpose convolution

Hello everyone , I meet you again , I'm your friend, Quan Jun .

When the forward convolution step is not 1 when （ It is often possible to 2）, The transpose convolution step is ：

The transposed convolution filling is determined by the convolution kernel and filling of the forward convolution ：

The kernel of transposed convolution is the same as that of forward convolution ：

The output formula of transpose convolution ：

Transpose convolution is also called micro step convolution (“ A little step ” Means ： The new step size is 1, The previous step size is 2, The sliding window processing of transpose convolution is compared with that of convolution “ Small ”.), It can be regarded as a kind of traditional convolution operation “ reverse ” Transmission process ; also , Transpose convolution is subject to “ positive ” Parameter constraint of convolution , That is, step length stride And zero padding （zero-padding）. Here's how stride=1 and padding=0、stride=1 and padding=1、stride=2 and padding=0、stride=2 and padding=1 Examples of convolution and transpose convolution .

conv_arithmetic： Click to open the link

Publisher ： Full stack programmer stack length , Reprint please indicate the source ：https://javaforall.cn/151923.html Link to the original text ：https://javaforall.cn