AdaIN-pytorch

PyTorch implementation of "Arbitrary Style Transfer in Real-time with Adaptive Instance Normalization" by Xun Huang, Serge Belongie. arxiv.

0. References

• irasin/Pytorch_AdaIN

1. Set up and train model

1.1 install requirements

pip install -r requitements.txt

1.2 get the datasets and unzip them to desired location

• COCO: https://cocodataset.org/#download
• WikiArt: https://www.kaggle.com/c/painter-by-numbers/data

1.3 Train model

• to start training: python main.py train path/to/coco path/to/wikiart [OPTIONS]

• to change training parameters and options: python main.py train --help

2. Style Transfer using pre-trained model

• to run inference: python main.py infer [OPTIONS]
• info on supported options: python main.py infer --help

3. Some Results

• All images were resized to 1024x1024 for inference. The 1024x1024 outputs are interpolated to the original size using bilinear interpolation.

Content Img	Style Img	Output