neural-network-renderer 0.1

A usefull CNN/DenseNet visualization tool

neural-network-renderer

This repository contains everything necessary to generated neural network visualisation.

The code generating the image is writen in Python. This code generates .texfile that is directly compiled and deleted once the results are available.

Example

Here is an example of the code to generate a simple convolutionnal-network representation:

from pathlib import Path
import sys

from network_layer.architecture import Architecture
from network_layer.layers import Input, Pool, Conv, Softmax, Dense, Spacer


def main():
    arch = Architecture(2 / 32)

    # input
    arch.add(Input("assets/input_def_crop.png", shape=[64, 64]))

    # first layer
    arch.add(Conv([64, 64, 32], s_filter=64, n_filter=32, to="(4,0,0)"))
    arch.add(Conv([62, 62, 32], s_filter=62, n_filter=32))
    arch.add(Pool([31, 31, 32]))

    arch.add(Spacer())

    # second layer
    arch.add(Conv([31, 31, 64], s_filter=31, n_filter=64))
    arch.add(Conv([29, 29, 64], s_filter=29, n_filter=64))
    arch.add(Pool([14, 14, 64]))

    arch.add(Spacer())

    # third layer
    arch.add(Conv([13, 13, 64], s_filter=14, n_filter=64))
    arch.add(Conv([12, 12, 64], s_filter=12, n_filter=64))
    arch.add(Pool([6, 6, 64]))

    arch.add(Spacer())

    # flatten
    arch.add(Dense(64, 64))

    # output
    arch.add(Spacer())
    arch.add(Softmax(5, 5))

    arch.generate(f"{Path(sys.argv[0]).stem}.tex")


if __name__ == "__main__":
    main()

The rendering is started with this command:

bash to_pdf.sh main

where main is the name of the python file.