# PyTorch implementation of StyleTransferGNN Paper: Learning Graph Neural Networks for Image Style Transfer, ECCV'22 ## For Easy Comparison Since our code currently needs at least 30GB GPU memory for inference, a zip file named "Results_For_Easy_Comparison.zip" has been uploaded for a easy comparison with our method in the paper, with our various results as well as the content and style images. Please feel free to download and use them in your paper. Thanks! ## Method Overview ![Architecture][1] [1]: https://files.osf.io/v1/resources/x5cez/providers/osfstorage/62d65298588bb93864b874f2?mode=render ## Dependencies Please kindly see requirements.txt in the zip file for more information about how to install the dependencies. A docker file is also provided for a easier environment setup. ## Usage Please kindly see README.md in the zip file. Thanks! # Notes Our code may currently look messy, due to the tight deadline of ECCV camera ready. I will reorganize the code in a few weeks and add more instructions on how to run the code. Thanks so much! # GPU Resources Please kindly note that our code needs at least 30 GB GPU memory. We are currently seeking for the solution to reduce the memory usage. Thanks! ## Citation If you find this code useful for your research, please kindly consider citing: ``` @inproceedings{jing2022learning, title={Learning Graph Neural Networks for Image Style Transfer}, author={Jing, Yongcheng and Mao, Yining and Yang, Yiding and Zhan, Yibing and Song, Mingli and Wang, Xinchao and Tao, Dacheng}, booktitle={ECCV}, year={2022} } ``` Thanks! ## License StyleTransferGNN is released under the MPL license. Only the academic use is allowed. Please see the LICENSE file for more information. ## Acknowledgements Our code is based on the wonderful work of [pytorch-AdaIN](https://github.com/naoto0804/pytorch-AdaIN) and [DPT](https://github.com/CASIA-IVA-Lab/DPT). We deeply appreciate their great codes!