CFCPalsy: Facial Image Synthesis with Cycle Cross-Fusion Diffusion Model for Facial Paralysis Individuals

This is the official repository for the paper CFCPalsy: Facial Image Synthesis with Cross-Fusion Cycle Diffusion Model for Facial Paralysis Individuals

This is an open-source project for facial expression transfer in facial palsy images, aimed at providing high-quality facial palsy expression synthesis methods. We approach the problem from the perspectives of feature extraction, feature fusion, and efficient model training. Based on a diffusion model, we implement multi-condition facial palsy expression image synthesis.

Environment preparation

pip install -r requirements.txt

Test CFCPalsy utilizing given sample images

You can quickly experience the facial palsy face synthesis results by first using the model we provide.

• Download the pretrained weights from the link
• Place the pretrained_models directory under ccfexp (same level as src)
• Run

cd ccfexp/src
bash syn.sh

Acknowledgement

This codebase is built based on DCFace. Thank you very much!