All convolutions inside of a dense block are ReLU-activated and use batch normalization. Channel-sensible concatenation is only attainable if the height and width dimensions of the data continue to be unchanged, so convolutions inside of a dense block are all of stride 1. Pooling levels are inserted between dense https://financefeeds.com/nibiru-evm-nibiru-v2-codebase-finalized-following-rigorous-security-audits/