3X3 convolution is achieved by matrix multiplication; A is a matrix of [W X H, 9], B is a matrix of [9, 1], and output will be a matrix of [W X H, 1], which could be converted into a [W, H] matrix.
kristinayige
Note: this actually explained that for a convolution action, each pixel is being touched/involved in calculations for filter_size^2 times.
3X3 convolution is achieved by matrix multiplication; A is a matrix of [W X H, 9], B is a matrix of [9, 1], and output will be a matrix of [W X H, 1], which could be converted into a [W, H] matrix.