All convolutions within a dense block are ReLU-activated and use batch normalization. Channel-wise concatenation is simply feasible if the peak and width dimensions of the information keep on being unchanged, so convolutions inside of a dense block are all of stride one. Pooling levels are inserted between dense blocks https://financefeeds.com/cftc-seeks-copyright-user-data-in-polymarket-probe/