The figure problem

[Reagan1311]

After reproducing your network, I found this figure in your paper may lead to some understanding, as each EDA module didn't take all preceding feature maps as input, so it's different from DenseNet.

[shaoyuanlo]

DenseNet also has several Dense Block, and there is a transition layer between every two adjacent blocks. In my work, EDA Blocks can be corresponding to Dense Blocks, and Downsampling Blocks can be corresponding to transition layers. Therefore, it seems nothing wrong in my paper. It is more precise to say "each EDA module takes all preceding feature maps in the current EDA Block as input". Thanks for your question!