This repository contains implementation of Neural/image Style Transfer based on paper "A Neural Algorithm of Artistic Style" by Gatys et. al.
Apart from the content loss and style loss as mentioned in the paper a new loss called "Continuity loss" is introduced by us in the implementation. This loss accounted for the differences on consecutive frames in the stylised image when working on a gif.
python3 neural_style_red.py --content_folder <content_folder> --style <style_file> --output_folder <output_folder>
<content_folder> must contain the each frame in order of the gif <output_folder> is the destination where the stylized frames are stored
A gif file named "output.gif" is also stored at the location of the "neural_style_red.py" file folder
For obtaining the results for a single image the following command can be used :
python3 neural_style1.py --content <content_file> --styles <style_file> --output <output_file>
Use --iterations to change the number of iterations Use --content-weight to change the weight of content of content image in the output image Use --style-weight to change the weight of style of style image in the output image
A pretrained VGG19 network is put in the the top level of this repository, or specify its location using the --network option.
The packages required for the code to work are :
- Tensorflow
- NumPy
- SciPy
- Pillow
- Imageio