search

tzutalin / dlib-android-app

:dragon: Android app to demo dlib-android(https://github.com/tzutalin/dlib-android). Use the prebuilt shared-lib built from dlib-android
Apache License 2.0
683 stars 246 forks source link

slow facial landmarks detection #9

Closed chrysantd closed 8 years ago

chrysantd commented 8 years ago

Thanks for your great work. It is very useful. However, when I tried demo app to detect facial landmarks, it took more than 10s to finish the whole process for only one picture. Is there any configurations I need to notice? Or it is the normal speed?

tzutalin commented 8 years ago

Hi, How did you profile? What is your image resolution?

Except to facial landmark, I also do HOG Opencv detection before doing facial landmark.

Regards, TzuTa lin

Thanks for your great work. It is very useful. However, when I tried demo app to detect facial landmarks, it took more than 10s to finish the whole process for only one picture. Is there any configurations I need to notice? Or it is the normal speed?

— You are receiving this because you are subscribed to this thread. Reply to this email directly or view it on GitHub https://github.com/tzutalin/dlib-android-app/issues/9

chrysantd commented 8 years ago

It is a 250*250 jpeg image. Another problem, I trained a 15-points model, and change the path in getFaceShapeModelPath() function, but it still uses the 68-point model...

tzutalin commented 8 years ago

Hi @chrysantd , After check, it takes 3~4 seconds to load model for the first time.

Then, usually, it takes 1~2 seconds detect face and landmark for 225x225 images.

First time:

01-02 21:26:23.286  5034  5232 I native  : I/jni_people_det.cpp:306 com_tzutalin_dlib_PeopleDet jniDLibHOGFaceDetect
..
01-02 21:26:28.427  5034  5232 I native  : I/jni_people_det.cpp:147 Dlib HOG face det size : 1
01-02 21:26:28.434  5034  5232 I native  : I/jni_people_det.cpp:153 face index:0number of parts: 68

Second time:

8.055  5034  5232 I native  : I/jni_people_det.cpp:306 com_tzutalin_dlib_PeopleDet jniDLibHOGFaceDetect
...
01-02 21:31:49.541  5034  5232 I native  : I/jni_people_det.cpp:147 Dlib HOG face det size : 1
01-02 21:31:49.548  5034  5232 I native  : I/jni_people_det.cpp:153 face index:0number of parts: 68

In general, it should load the model according getFaceShapeModelPath(). Can you share with me about how you do that? I can help you to verify your model.

tzutalin commented 8 years ago

@chrysantd , Could you share your model to this repository?

bpkrishnan commented 8 years ago

The default shape_predictor_68_face_landmarks.dat itself is taking 10sec to load initaly. What to do?

gerardsimons commented 7 years ago

I am also facing this latency issue. Is it possible to load it before the camera activity and keeping the model in memory some how?

Seetaiah74 commented 7 years ago

Hi, shape_predictor_68_face_landmarks.dat gallery image its taking 10-15 secs, camera activity facing issue running background threads to capture and convert, will u share me the perfect procedure for live camera preview frames?

WaqarAhmadGit commented 7 years ago

Hello I got 3 points of mouth and now i want to change mouth means lips colors or adding lips image so user can see how other colors looks on his / her lips kindly guide me if any one have idea your response will be much appreciated. Thanks