can you tell me how you optimized the dlib ?

[9crk]

can you tell me how you optimized the dlib ?
will you provide the source code?

[gicheonkang]

Hi, here are the concept applied in this project 1. multi-threaded dlib 2. lower number of pyramidal layer I will send you a source code asap :) 2017. 5. 25. 오후 12:21에 "周华" <notifications@github.com>님이 작성:

…

can you tell me how you optimized the dlib ? will you provide the source code? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#1>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AWTBITHXRjbJuTSJZqlWc987yyrK3f0zks5r9PO-gaJpZM4Nl6ra> .

[9crk]

lower number of pyramidal layer

does that means doing face detection?

now my porblem is mainly on landmark predictor part.it's quiet slow.about 150-200ms.

I don't know why..

[9crk]

davisking/dlib#604
this is the process I did.
I use the code of dlib version 19.1 because I cant get through the version 19.4 on simd part's compile.

[9crk]

I found the solution,...I should build the project in release mode..

[gicheonkang]

Good work ! Actually, predicting landmark points is not a time-consuming work because dlib supports it. If you want to make a fast face predictor, you have to reduce the time cost of detecting faces. That’s a critical part :) There are many ways to optimize the application, such as resizing the image, making a grayscale image, using a NEON architecture, Compiling release mode ... You did it on Android platform ?

…

2017. 5. 25. 오후 4:41, 周华 ***@***.***> 작성: now it turns 15ms for landmark. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#1 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AWTBITE3x5rsIbtuqqR0BBsNswM47obNks5r9TCygaJpZM4Nl6ra>.

[9crk]

yes,on Android device.
did you integrated the NEON support ?
I use the 19.1 source code of dlib, because I cant compile the 19.4 through.the simd fils always get errors.

I use the opencv to do face detection, I can optimize it to under 10ms and still usable.

[9crk]

did you eat 粽子 these days btw? :-)

[gicheonkang]

Hi, 周华 I’m looking for the ways to port android-version(NEON support) @fastfastball said that he did it ! Here is the repo https://github.com/fastfastball/dlib_for_arm <https://github.com/fastfastball/dlib_for_arm> But, when I replace original one(19.1) with new one(19.4, simd). Error occurs. Maybe it needs to be setting the Android.mk, Application.mk file Test it ! If you successfully compiled, please let me know asap :) Thank you.

…

2017. 5. 31. 오후 7:12, 周华 ***@***.***> 작성: did you eat 粽子 these days btw? :-) — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#1 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AWTBIX9Efm0zeNeG9mftF47pr89YS-9jks5r_Tz9gaJpZM4Nl6ra>.