All convolutions inside a dense block are ReLU-activated and use batch normalization. Channel-smart concatenation is simply attainable if the peak and width dimensions of the information keep on being unchanged, so convolutions in a very dense block are all of stride 1. Pooling levels are inserted between dense blocks https://financefeeds.com/heres-why-dogecoin-doge-may-never-hit-50-and-under-1-doge-alternative-set-to-achieve-it-by-2026/