Icon Generator

How to use:

Open IconGenerator.ipynb with Jupyter Notebook.

Tensorflow and Keras should be installed, as well as a range of standard libraries like numpy. The tensorflow docker container already contains all the necessary libraries.

Training Dataset

The easiest way to get a dataset for training is to download the LLD dataset: https://data.vision.ee.ethz.ch/sagea/lld/

Alternatively, icons can be downloaded seperately and loaded from a folder, the process is described in the iPython notebook.

Pre-Trained Weights

Weights for the results of the improved convolutional GAN and the ResNet WGAN can be downloaded here:

https://drive.google.com/open?id=1FFKdxTGTTUdLZQOeoWkqFb9AegGm3dSF

Commands for the Tensorflow Docker Container

Command for jupyter docker container with current directory as connected drive:

docker run -v $(pwd):/tf/icon-gan -it --rm -p 8888:8888 tensorflow/tensorflow:latest-py3-jupyter

Command with Nvidia GPU enabled, recommended for training:

docker run --runtime=nvidia -v $(pwd):/tf/icon-gan -it --rm -p 8888:8888 tensorflow/tensorflow:latest-gpu-py3-jupyter