readme.md

Embedding and Keypoints Extractor for Real and Anime Face

FaceEKE is a specialized method for extracting face embeddings and keypoints (KPS) from input character images. It is based on the open-source Insightface model and can handle over 90% of face embedding extraction tasks. For more challenging cases, such as detecting anime faces, we use the Anime Face Detector model for landmark detection. This approach can ultimately be adapted for embedding extraction with antelopev2 model.

FaceEKE provides a solution for embedding extraction in downstream tasks such as face alignment, face swapping, and FaceID generation, especially for anime faces.

---

How to run

git clone https://github.com/ykj467422034/FaceEKE.git
cd face_embedding_extractor
pip install -r requirements.txt
python extract_face_embedding.py

Demos

Compare with original Insightface and Anime Face Detector

Insightface	Ours

Demos of our method

Image	face and kps

Acknowledgements

FaceEKE is developed by KUNLUN Skywork AIStory team, all copyright reserved.

Thanks to insightface for provide face embedding model.

Thanks to anime-face-detector for anime face detect.