pytorch-fns

Environment

// If env already created
source activate pytorch 

// If not, let's create it
conda create pytorch python=3.6
conda install pytorch=0.4.2
pip install -r requirements.txt

Train

rm -f ~/Documents/data/models/pytorch-checkpoints/*.pth \
&& \
rm -f ~/Documents/data/images/pytorch/*.jpg \
&& \
python neural_style/neural_style.py train \
--dataset ~/Documents/data/training/images \
--style-image ~/Documents/data/images/styles/beer.jpg \
--batch-size 2 \
--style-size 800 \
--style-weight 1.2e10 \
--content-weight 1e5 \
--alpha 0.3 \
--checkpoint-model-dir ~/Documents/data/models/pytorch-checkpoints \
--checkpoint-interval 1000 \
--save-model-dir ~/Documents/data/models \
--log-interval 10 \
--cuda 1

Export to CoreML

python ./converter.py  \
--pth_model ~/Documents/data/models/pytorch-checkpoints/checkpoint_41000.pth \
--output ~/Documents/data/models/mlmodels/beer_fast_2.mlmodel \
--alpha 0.3

Stylize test

python neural_style/neural_style.py eval \
--model ~/Documents/data/models/pytorch-checkpoints/checkpoint_5000.pth \
--content-image ~/Documents/data/images/test.jpg \
--output-image ~/Documents/data/images/stylized-test.jpg \
--alpha 0.3 \
--cuda 0

python ./src/main.py test \
--content-image ~/Documents/data/images/test.jpg \
--output-image ~/Documents/data/images/stylized-test.jpg \
--model ~/Documents/data/models/meta.pth \
--cuda 1