"Neural Network and deep learning" convolution and deconvolution

1. Convolution vs. deconvolution

The above figure illustrates the process of convolution of the core deconvolution, defining the input matrix I (4x4), the convolution kernel is K (3x3), the output matrix is O (2x2): The process of convolution is: Conv (i,w) =o deconvolution over called: Deconv (W,o) =i (need to the O at this time Extension padding) 2. Step and overlap

When the step size of the convolution core movement (stride) is smaller than the side length of the convolution nucleus (usually a square row), the overlap occurs when the convolution nucleus overlaps with the original input matrix (overlap), and the Step (stride) of the convolution nucleus moves in accordance with the edge appearance of the convolution nucleus.

4x4 Input matrix