smnai / style_transfer

Style transfer in PyTorch à la Gatys et al. 2015

PyTorch Style Transfer

More or less replicating Gatys et al. 2015 but not, like, exactly

Usage

Install requirements with:

pip install -r requirements.txt

Put the content and style images in the img folder and run

./src/style.py [CONTENT IMAGE NAME] [STYLE IMAGE NAME] 

with additional flags:

[-g STEPS] [-s IMAGE_SIZE] [-l NUM_STYLE_LAYERS] 
[-r RANDOM INITIAL IMAGE] [-a ALPHA] [-b BETA] [-n NORMALIZE INPUT]

To get parameter descriptions and defaults, run:

./src/style.py --help

Just FYI, there's plenty of better Pytorch implementations of Gatys et al. 2015 online, this one is not very well optimized or tested: I wrote it over a weekend to test my understanding of style transfer.

Gallery