neural-style-transfer

This repository contains an implementation of the work presented in the paper A Neural Algorithm of Artistic Style by Leon A. Gatys, Alexander S. Ecker, Matthias Bethge. Based on a style and content image, a new target image is created, which contains the content of the content image and the style of the style image. The content and style information are extracted from convolutional layers of a VGG-19 model trained on ImageNet.

Installation

The easiest way to get the provided code running is by creating a python environment based on the provided environment.yml file.

$ conda env create -f environment.yml

This creates a conda environment called neural-style-transfer, which shall be activated prior to use of the command line tool.

$ conda activate neural-style-transfer

If the conda environment already exists, it can be activated with the following command.

$ conda env update -f environment.yml

Usage

Usage: neural-style-transfer.py [OPTIONS] STYLE_IMAGE CONTENT_IMAGE
                                TARGET_IMAGE

Options:
  --epochs INTEGER  Number of epochs
  --alpha INTEGER   Content weight
  --beta FLOAT      Style weight
  --help            Show this message and exit

Examples

The following example was created with --epochs 5000.

Style	Content	Result

License

neural-style-transfer is Copyright © 2019 Alexander Stante. It is free software, and may be redistributed under the terms specified in the LICENSE file.