1. Introduction:
tradtional convolutional networks with L layer with L networks: one between each layer is subsequent layer. Densely Connected Convolutional Networks has L*(L+1)/2 layers.
2. Structure:
Base on feed-forward nature, each layer obtain addtional input from all proceeding layers and passed all feature map to subsequent layers.
3. Highlight :
A possibly counter-intuitive effect os this pattern is that it requires few paraments than tradtional Convolutional Networks, as there no need to relearn redundant feature-maps.