All convolutions within a dense block are ReLU-activated and use batch normalization. Channel-sensible concatenation is simply feasible if the height and width Proportions of the information stay unchanged, so convolutions inside a dense block are all of stride one. Pooling layers are inserted between dense blocks for additional dimensionality https://financefeeds.com/best-meme-coins-to-buy-now-top-picks-for-2025-supercycle/