All convolutions in a very dense block are ReLU-activated and use batch normalization. Channel-sensible concatenation is only possible if the height and width dimensions of the information continue to be unchanged, so convolutions in a very dense block are all of stride one. Pooling levels are i