All convolutions inside a dense block are ReLU-activated and use batch normalization. Channel-smart concatenation is barely possible if the height and width dimensions of the data continue to be unchanged, so convolutions within a dense block are all of stride one. Pooling levels are inserted between dense blocks for https://financefeeds.com/doge-witnesses-big-move-as-shib-token-burn-rate-skyrockets-new-altcoin-projected-to-hit-1-as-launch-date-nears/