CNN: full convolution, same convolution, valid convolution

Such as the title: Convolutional neural networks can be divided into full convolution, same convolution, and full convolution.

Difference: The original image will become larger after full convolution, the same convolution size will remain unchanged, and the valid convolution size will shrink. The application scenario is unreasonable, see the following for details:

1. Full convolution

 

The orange part in the above picture is the size of the original image is 6x7, the convolution kernel is 3x3, and the step size is one for convolution. The full convolution starts from the intersection of the convolution kernel and the image , and the result is an 8*9 image, and the image is enlarged. Mainly used for upsampling, also known as enlarged image, image interpolation ; the main purpose is to enlarge the original image so that it can be displayed on a higher resolution display device.

Two, same convolution

 

 The same convolution is convolved from the case where the center of the convolution kernel intersects the corner k of the original image. When the step size is one, the original image and the output size are the same.

 As shown in the figure above: Convolution starts when the convolution kernel and the image completely intersect. The original image is 6x7, and the convolution result is 4x5, and the size is reduced.

There are two main purposes: 1. Make the image conform to the size of the display area; 2. Generate a thumbnail of the corresponding image;