Awesome

We plan to create a very interesting demo by combining Segment Anything and a series of style transfer models! We will continue to improve it and create more interesting demos. Interesting ideas, results, and contributions are warmly welcome!

[Arxiv]

Demo

Different contents can specify different styles in a style image. Left click to select an area and right click to exclude an area.
Middle click and drag your mouse to specify a bounding box.
You can also draw a contour with mouse and specify a style region for it.

Installation

python -m pip install torch
python -m pip install -e segment_anything
python -m pip install opencv-python

The code is tested on the environment with Ubuntu 18.04, python 3.6.9, torch 1.8.1+cuda10.2, opencv-python 4.5.5, and a 2080Ti GPU.

Get Started

Clone this repo:

git clone https://github.com/Huage001/Transfer-Any-Style.git
cd Transfer-Any-Style

Download the model checkpoint of Segment Anything:

wget https://dl.fbaipublicfiles.com/segment_anything/sam_vit_h_4b8939.pth
mv sam_vit_h_4b8939.pth segment-anything/

Download the model checkpoint of AdaAttN from here and unzip it to directory of this repo:
```
mv [DOWNLOAD_PATH]/ckpt.zip .
unzip ckpt.zip
rm ckpt.zip
```
Run the following command:
```
python transfer_any_style.py \
--content_path CONTENT_PATH \
--style_path STYLE_PATH \
--resize
```
Follow the instruction printed on the console to run the interactive demo.
Full usage:
- Command: python transfer_any_style.py [-h] --content_path CONTENT_PATH --style_path STYLE_PATH [--output_dir OUTPUT_DIR] [--resize] [--keep_ratio]
- Arguments:
  - --content_path: Path to a single content img
  - --style_path: Path to a single style img
  - --output_dir: Output path
  - --resize: Whether resize images to the 512 scale, which is the training resolution of the model and may yield better performance
  - --keep_ratio: Whether keep the aspect ratio of original images while resizing

Future Work

Support more style images at a time.
Integrate with more state-of-the-art style transfer methods.
More user-friendly and stable user interface.
...

:cupid: Acknowledgement

Citation

If you find this project helpful for your research, please consider citing the following BibTeX entry.


@article{liu2023any,
  title={Any-to-Any Style Transfer},
  author={Liu, Songhua and Ye, Jingwen and Wang, Xinchao},
  journal={arXiv preprint arXiv:2304.09728},
  year={2023}
}

@article{kirillov2023segany,
    title={Segment Anything}, 
    author={Kirillov, Alexander and Mintun, Eric and Ravi, Nikhila and Mao, Hanzi and Rolland, Chloe and Gustafson, Laura and Xiao, Tete and Whitehead, Spencer and Berg, Alexander C. and Lo, Wan-Yen and Doll{\'a}r, Piotr and Girshick, Ross},
    journal={arXiv:2304.02643},
    year={2023}
}

@inproceedings{liu2021adaattn,
  title={AdaAttN: Revisit Attention Mechanism in Arbitrary Neural Style Transfer},
  author={Liu, Songhua and Lin, Tianwei and He, Dongliang and Li, Fu and Wang, Meiling and Li, Xin and Sun, Zhengxing and Li, Qian and Ding, Errui},
  booktitle={Proceedings of the IEEE International Conference on Computer Vision},
  year={2021}
}

@article{yu2023inpaint,
  title={Inpaint Anything: Segment Anything Meets Image Inpainting},
  author={Yu, Tao and Feng, Runseng and Feng, Ruoyu and Liu, Jinming and Jin, Xin and Zeng, Wenjun and Chen, Zhibo},
  journal={arXiv preprint arXiv:2304.06790},
  year={2023}
}