It is very simple, and powerful cnn architecture.

This papers main idea is this - Don't discard previous feature!

Traditional CNN extract image feature from CNN layers, but it doens't use the CNN layer's features for classification.
In contrast, After Resnet, many CNN Architecture using skip connection(Resnet) and elementwise sum(Googlenet).

This paper is one of them, but the difference of this architecture doesn't excecute any Calculation,
just concatenate the previous layer to next layer

If you want more information,
you should check here (https://arxiv.org/abs/1608.06993)

You need Tensorlofw 2.x and numpy

You can install Tensorflow and numpy below code

pip install tensorflow

If you want to use gpu,

pip install tensorflow-gpu

And you also need numpy

pip install numpy

And you clone this repository.

git clone https://github.com/yw0nam/DenseNet

Finally run train.py

cd DenseNet
python train.py

That's all!
If you want some insight of this implementation, please look the model.ipynb