Facexlib Versions Save

🚀 Long time no see ☄️

✨ Highlights ✅ support change model root path to save models, so that facexlib does not need to save the models in the site-pacakge path. ✅ support on CPU mode for FaceRestoreHelper ✅ add parsenet to face_restoration_helper

🚀 Have a nice day

:sparkles: Highlights

• [Enhancement] We have updated face_restoration_helper.py. It now uses cv2.LMEDS method in the cv2.estimateAffinePartial2D, so that it can produce the same result as that of skimage transform.
• We add a face parsing net for low-quality face inputs.

This release also stores the following pre-trained models ;-)

1. parsing_parsenet.pth

🚀 Have a nice day!

✨ Highlights

Fix bugs in FaceRestoreHelper:

1. Fix the inaccurate paste back (back alignment) in FaceRestoreHelper: https://github.com/xinntao/facexlib/commit/a3395607634bb1ab9455621eaedb41ceab346466
2. Remove faces with too small eye distance, such as side faces or too small faces: https://github.com/xinntao/facexlib/commit/24493fc888acb03ca627ed07cd75376ab56c85bc, the example is here: https://github.com/xinntao/Real-ESRGAN/issues/72

Enhancement in FaceRestoreHelper

1. Support alpha channel and gray images: https://github.com/xinntao/facexlib/commit/3c06885f8b12b47863a2be4c258397fc1ef23974

🚀

:sparkles: Highlights

We have updated a lot!

1. face_restoration_helper supports pad_blur for multiple faces
2. Add face matting
3. Add face parsing
4. Add face head pose estimation
5. Add assessment

This release also storing the following pre-trained models ;-)

1. assessment_hyperIQA.pth
2. matting_modnet_portrait.pth
3. parsing_bisenet.pth
4. headpose_hopenet.pth

🚀

:sparkles: Highlights

Wow, we have a new lib -- facexlib.

This release is mainly for storing the following pre-trained models ;-)

1. alignment_WFLW_4HG.pth
2. detection_mobilenet0.25_Final.pth
3. detection_Resnet50_Final.pth
4. recognition_arcface_ir_se50.pth