Too many false positive

[JamesHowlettLA]

Many false positive if confidence=97,
many false negative if confidence=99, both false positive and false negative if confidence = 98.

I use fdt_demo to test.

[SunnyFog]

It detected the primary face reasonably well, but there are a lot of false positives here.

Image 960 x 540 17 faces detected. face_rect=[ 91x 91 @( 639, 166)], confidence=99, angle=0 face_rect=[ 19x 19 @( 354, 500)], confidence=89, angle=0 face_rect=[ 23x 23 @( 416, 500)], confidence=87, angle=0 face_rect=[ 23x 23 @( 880, 500)], confidence=87, angle=0 face_rect=[ 25x 25 @( 557, 497)], confidence=86, angle=0 face_rect=[ 29x 29 @( 557, 479)], confidence=63, angle=0 face_rect=[ 23x 23 @( 579, 501)], confidence=30, angle=0 face_rect=[ 23x 23 @( 899, 501)], confidence=30, angle=0 face_rect=[ 21x 21 @( 607, 160)], confidence=29, angle=0 face_rect=[ 23x 23 @( 940, 496)], confidence=26, angle=0 face_rect=[ 9x 9 @( 0, 506)], confidence=25, angle=0 face_rect=[ 17x 17 @( 132, 28)], confidence=25, angle=0 face_rect=[ 22x 22 @( 922, 500)], confidence=25, angle=0 face_rect=[ 24x 24 @( 840, 16)], confidence=25, angle=0 face_rect=[ 24x 24 @( 96, 499)], confidence=25, angle=0 face_rect=[ 27x 27 @( 805, 496)], confidence=25, angle=0 face_rect=[ 26x 26 @( 791, 497)], confidence=25, angle=0

[ShiqiYu]

You can remove the results with lower confidence. I just want users can choose a appropriate threshold according to different applications.

On Thu, Dec 5, 2019 at 9:11 AM SunnyFog notifications@github.com wrote:

![result_cnn_screenshot_04 12 2019](https://user-

images.githubusercontent.com/58533847/70194838-b4b86b80-16b8-11ea-8aaf-b736e88202eb.png )

It detected the primary face reasonably well, but there are a lot of false positives here.

Image 960 x 540 17 faces detected. face_rect=[ 91x 91 @( 639, 166)], confidence=99, angle=0 face_rect=[ 19x 19 @( 354, 500)], confidence=89, angle=0 face_rect=[ 23x 23 @( 416, 500)], confidence=87, angle=0 face_rect=[ 23x 23 @( 880, 500)], confidence=87, angle=0 face_rect=[ 25x 25 @( 557, 497)], confidence=86, angle=0 face_rect=[ 29x 29 @( 557, 479)], confidence=63, angle=0 face_rect=[ 23x 23 @( 579, 501)], confidence=30, angle=0 face_rect=[ 23x 23 @( 899, 501)], confidence=30, angle=0 face_rect=[ 21x 21 @( 607, 160)], confidence=29, angle=0 face_rect=[ 23x 23 @( 940, 496)], confidence=26, angle=0 face_rect=[ 9x 9 @( 0, 506)], confidence=25, angle=0 face_rect=[ 17x 17 @( 132, 28)], confidence=25, angle=0 face_rect=[ 22x 22 @( 922, 500)], confidence=25, angle=0 face_rect=[ 24x 24 @( 840, 16)], confidence=25, angle=0 face_rect=[ 24x 24 @( 96, 499)], confidence=25, angle=0 face_rect=[ 27x 27 @( 805, 496)], confidence=25, angle=0 face_rect=[ 26x 26 @( 791, 497)], confidence=25, angle=0

— 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/185?email_source=notifications&email_token=ABWR4HLUWAZRVORURFETFXLQXBIMHA5CNFSM4I6CH34KYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEF7DUXA#issuecomment-561920604, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABWR4HPGLXUESXBVJ7BPXRLQXBIMHANCNFSM4I6CH34A .