xiaowenhe / centerface

CenterFace

Introduce

CenterFace(size of 7.3MB, and AP 88.0% on the WIDER FACE hard set) is a practical anchor-free face detector for edge devices. CenterFace-small achieving AP equal to CenterFace while model size is only 2.3M.

Recent Update

• 2019.09.13 CenterFace is released.

Environment

• OpenCV 4.1.0
• Numpy
• Python3.6+

Accuracy

• Results on val set of WIDER FACE:

Model Version	Easy Set	Medium Set	Hard Set
RetinaFace-mnet	0.896	0.871	0.681
LFFD-v1	0.910	0.881	0.780
LFFD-v2	0.837	0.835	0.729
CenterFace	0.939	0.929	0.880
CenterFace-small	0.931	0.924	0.870

• RetinaFace-mnet is short for RetinaFace-MobileNet-0.25 from excellent work insightface)
• LFFD-v1 is from prefect work LFFD
• CenterFace/CenterFace-small evaluation is under MULTI-SCALE, FLIP.

• Results on FDDB:

Model Version	Disc ROC curves score
RetinaFace-mnet	96.0@1000
LFFD-v1	97.3@1000
LFFD-v2	97.2@1000
CenterFace	97.9@1000
CenterFace-small	98.1@1000

Inference Latency

• Latency on NVIDIA RTX 2080TI:

Resolution->	640×480	1280×720(704)	1920×1080(1056)
RetinaFace-mnet	5.40ms	6.31ms	10.26ms
LFFD-v1	7.24ms	14.58ms	28.36ms
CenterFace	5.5ms	6.4ms	8.7ms
CenterFace-small	4.4ms	5.7ms	7.3ms

Joint face detection and alignment By CenterFace

We also try the mutil-task Center Face to joint face detection and alignment. The model is releasing soon.

Results

Discussion

Welcome to join in QQ Group(912759877) for more discussion, including but not limited to face detection, face anti-spoofing and so on.