Welcome to the official implementation of D2Styler, which has been accepted at the International Conference on Pattern Recognition (ICPR 2024).
"D2Styler: Advancing Arbitrary Style Transfer with Discrete Diffusion Methods" introduces a novel framework for style transfer called D2Styler. Leveraging VQ-GANs and discrete diffusion, this method aims to improve the quality and stability of style transfer, addressing common issues like mode-collapse and over/under-stylization. By using Adaptive Instance Normalization (AdaIN) features, D2Styler facilitates effective style transfer between images. Experimental results show that D2Styler outperforms twelve existing methods on various metrics, producing high-quality, visually appealing images. The method uses images from the WikiArt and COCO datasets. The model's architecture and its qualitative results are showcased below. The model will be available on HuggingFace 🤗, where you can download it for inference or fine-tuning.
To get started with D2Styler, follow the steps below to install the necessary dependencies:
-
Clone the repository:
git clone https://github.com/yourusername/D2Styler.git cd D2Styler -
Install the dependencies:
pip install -e ".[torch]" pip install -e .[all,dev,notebooks]
We welcome contributions to D2Styler! If you have any ideas for improvements or find any issues, please feel free to open an issue or submit a pull request.
For more details, please refer to our paper and our repository on HuggingFace.