Closed dassima closed 7 years ago
It's not trained to recognize side view faces, so it probably wouldn't work well for that use case unfortunately.
@ageitgey How would we go about trying to train it for this?
I'm new to python and the whole machine learning side of things and I'm just curious if there could ever be a feature for this or if you know of any other libraries for this?
If you just want to detect side-view faces, you could create a bunch of training data and train an object detector like this: http://blog.dlib.net/2016/10/easily-create-high-quality-object.html. That wouldn't be too hard to do. That's the same way the frontal face detector works.
If you want facial feature detection (eyes, nose, etc) and face recognition with side view faces, it would be a lot more work.
Thanks for getting back to me and the link to the tutorial 👍
@ageitgey, the face_recognition is great! But I also want to be able to recognise side faces. I have implemented the side face detection and could also do the side face recognition using a neural network separately from this model. But your model works so well that I want to be able to integrate side face recognition with the existing frontal recognition. Right now I'm stuck because I cannot find much information on how the encodings are generated from the front faces and if you could suggest me something it would be great.
For detection (but not recognition) of side faces, this lib works very well and is much faster than face_recognition https://github.com/arunponnusamy/cvlib
Hi all! I'm running into issues with dlib and side face detections. For some reason, even for side faces,m dlibs returns all 68 landmarks, and then my cnn tries to recognize this face. I am setting it up, so if landmarks < 45, to ignore detections, but since everything is being returned as 68, my system tries to recognize, and fails miserably. Is this a dlib issue? or is it something I'm doing?? Really, all I want to do is dlib to return back an accurate number of landmarks detected. Can anyone comment? ps. not looking for recognition, only detection
@7thstorm I am trying to build a side face detector myself. Do you have any code you can share and I can try to see why that is occuring.
@ageitgey it does work on side view face but only in case of detection. I have performed it. But can you please help if there is any way we can train it to RECOGNIZE side view faces or is there any way that we collect some images in different position with angle lets say about 30 degree. It is possible ?
I think the solution is to train 2 different detectors. Maybe one cannot detect both frontal and side/profile. And then run the image through the 2 networks. Here are training parameters for another dlib model. https://github.com/Luca96/dlib-minified-models/tree/master/face_landmarks#training-parameters
@fadi212 https://github.com/ageitgey/face_recognition/issues/421#issuecomment-524605899 do you want to work on this together? I have unlimited free access to GTX 1080 GPU for this.
@nyck33 did you succeed in solving this?
No, my contract ended so I never finished. It shouldn't be that hard to train from scratch though. Just time consuming if you have to annotate your own dataset.
Get Outlook for Androidhttps://aka.ms/ghei36
From: Shalom John notifications@github.com Sent: Wednesday, December 11, 2019 1:53:47 AM To: ageitgey/face_recognition face_recognition@noreply.github.com Cc: Toyoda Kim, Nobutaka T nobutaka@gatech.edu; Mention mention@noreply.github.com Subject: Re: [ageitgey/face_recognition] Side view faces (#208)
@nyck33https://github.com/nyck33 did you succeed in solving this?
— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://github.com/ageitgey/face_recognition/issues/208?email_source=notifications&email_token=AGAFZKP5S7I46W6H44LCRYLQYC2CXA5CNFSM4D7KZQS2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEGSQ2LY#issuecomment-564464943, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AGAFZKJO4VVMYTY7ST7UCELQYC2CXANCNFSM4D7KZQSQ.
Does this algorithm detect side view faces in images too? I am looking for one to implement for TP-GAN.(Beyond face Rotation Paper)