Pytorch Adain Style

PyTorch Implementation of the paper "Arbitrary Style Transfer in Real-time with Adaptive Instance Normalization"

Notice: This repositories is simple implementation for self taught

Environment

• Python >= 3.8
• Pytorch >= 1.8

pip install -r requirements.txt

Dataset

• Content images: COCO
• Style images: Wikiart

Data structure

-project_name/
    -contents/
        -*.png(or others format)
    -styles/
        -*.png(or others format)
        
# for example
# The number of content images must be consistent with the number of style images
-Adain/
    -contents/
        -00001.png
        -00002.png
        -00003.png
    -styles/
        -00004.png
        -00005.png
        -00006.png

Training

python train.py --train-content contents --train-style styles --gpu 0,1 --workers 8 --epochs 100 --batch-size 16

Test

• Pretrained weights: weights

# source is directory
python test.py --weights epoch_100.pth --content contents_path --style styles_path

# source is file
python test.py --weights epoch_100.pth --content 0000.1.png --style 0000.2.png

Reference

• Original implementation
• irasin's implementation