search

1adrianb / face-alignment

:fire: 2D and 3D Face alignment library build using pytorch
https://www.adrianbulat.com
BSD 3-Clause "New" or "Revised" License
7.12k stars 1.35k forks source link

dlib's CNN face detection model performs worse than one from face-recognition-models #96

Closed jeongyoonlee closed 6 years ago

jeongyoonlee commented 6 years ago

Currently the code downloads dlib's CNN face detection model from https://www.adrianbulat.com/downloads/dlib/mmod_human_face_detector.dat.

Although it is supposed to be the same as one from face_recognition_models at https://github.com/ageitgey/face_recognition_models/blob/master/face_recognition_models/models/mmod_human_face_detector.dat, the one downloaded by this repo performs a lot worse than the one from face_recognition_models.

When tested with 1,000 images with one face in each image, face-alignment failed to detect faces from 363 images while face_recognition failed to detect faces from only 175 images.

Could you update mmod_human_face_detecctor.dat with latest version or the one from face_recognition_models?

1adrianb commented 6 years ago

Hi @jeongyoonlee,

Thanks a lot for the finding! I have replaced the model with the new one as suggested. However, for best results I recomend something like: https://github.com/clcarwin/SFD_pytorch

Thanks a lot!

jeongyoonlee commented 6 years ago

You're welcome and thanks for the recommendation, @1adrianb. I will check it out. :)