In the light of feature distributions: Moment matching for Neural Style Transfer (CVPR 2021)

This repository provides code to recreate results presented in In the light of feature distributions: Moment matching for Neural Style Transfer.

For more information, please see the project website and make sure to check out our medium blog post here

---

Contact

If you have any questions, please let me know

Instructions

Running neural style transfer with Central Moment Discrepancy is as easy as running

python main.py --c_img ./path/to/content.jpg --s_img ./path/to/style.jpg

You have the following command line arguments to change to your needs:

  --c_img         The content image that is being stylized.
  --s_img         The style image
  --epsilon       Iterative optimization is stopped if delta value of 
                  moving average loss is smaller than this value.
  --max_iter      Maximum iterations if epsilon is not surpassed
  --alpha         Convex interpolation of style and content loss 
                  (should be set high > 0.9 since we start with content as target)
  --lr            Learning rate of Adam optimizer
  --im_size       Output image size. Can either be single integer for keeping aspect ratio or tuple.

Citations

@article{kalischek2021light,
      title={In the light of feature distributions: moment matching for Neural Style Transfer}, 
      author={Nikolai Kalischek and Jan Dirk Wegner and Konrad Schindler},
      year={2021},
      eprint={2103.07208},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}