The goal of this project is to have a mirror that display the styled transfer image of the reflection in real time.
This work will be based on this paper :
Arbitrary Style Transfer with Style-Attentional Networks
Also took some code part from this repo: https://github.com/mumair5393/Style-Transfer-with-Style-Attentional-Networks
To train with the same data as the pretrain network, or to train more,
sh download_utils.sh
and download the pretrain weight at this adresse : https://drive.google.com/file/d/1Yd6FSHbEDzzZB8on9xJe7dSC8omOG6OE/view?usp=sharing
Put this weight on the model_save
folder.
Then unzip wikiart.zip
and train2017.zip
.
Move the images to the folders, respectively, style
and content
Finally use the train.ipynb
You can see some results here :
and a video of the style transfer mirror here