search

ShiqiYu / libfacedetection

An open source library for face detection in images. The face detection speed can reach 1000FPS.
Other
12.27k stars 3.05k forks source link

Is there any plan to open source tranditional face detection approach #98

Closed chibai closed 5 years ago

chibai commented 5 years ago

Dear Professor Yu: I've been watched the project for over 3 years. And I'm really happy and grateful that you can open source your code to make great contribution of the community. As far as I knew, at the very beginning of the project, you are actually using pure c with traditional approach (perhaps V-J approach?) for face detection, which I think I can really learn a lot from the pure c code, cause it's so fast that there must be many speed-up tricks. I wish you could open source the traditional-approach face detection code. It really doesn't matter the dl approache is faster than traditional approach, I'm just so respecting the tricks of the code may used,. Best Wishes!

ShiqiYu commented 5 years ago

Sorry I cannot release the AdaBoost based face detection algorithm because of some issues.

On Fri, Mar 15, 2019 at 7:14 PM chibai notifications@github.com wrote:

Dear Professor Yu: I've been watched the project for over 3 years. And I'm really happy and grateful that you can open source your code to make great contribution of the community. As far as I knew, at the very beginning of the project, you are actually using pure c with traditional approach (perhaps V-J approach?) for face detection, which I think I can really learn a lot from the pure c code, cause it's so fast that there must be many speed-up tricks. I wish you could open source the traditional-approach face detection code. It really doesn't matter the dl approache is faster than traditional approach, I'm just so respecting the tricks of the code may used,. Best Wishes!

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/ShiqiYu/libfacedetection/issues/98, or mute the thread https://github.com/notifications/unsubscribe-auth/AG0eHSqS5NlqvjbDz0NEkEz--Led7EMLks5vW4CWgaJpZM4b2PNe .

jamessmith90 commented 5 years ago

@chibai Don't give him any credit he has just copied the LBP algorithm which works on integers that is why it works fast. The real accuracy comes from the dataset and the traditional approach is not accurate enough to be used in production environment. Just train the wider face on MTCNN or on Yolov3-Tiny you will be surprised how good results you will get.

ShiqiYu commented 5 years ago

@chibai Don't give him any credit he has just copied the LBP algorithm which works on integers that is why it works fast. The real accuracy comes from the dataset and the traditional approach is not accurate enough to be used in production environment. Just train the wider face on MTCNN or on Yolov3-Tiny you will be surprised how good results you will get.

The last part of your comments is partly correct.