The output in the convolutional layer is generally handed from the ReLU activation function to bring non-linearity to your model. It requires the aspect map and replaces all the adverse values with zero. A VGG-block experienced lots of 3x3 convolutions padded by 1 to keep the size of output https://financefeeds.com/solana-and-xrp-etfs-could-bring-in-billions-say-jpmorgan-analysts-rollblock-ecosystem-sees-huge-user-growth/