人脸检测最新模型对比

算法名称	开源程度	热门程度	性能	准确率	训练集	框架基础
HyperFace	√	较热门	The Hyperface method was tested on a machine with 8 cores and GTX TITAN-X gpu. The overall time taken to perform all the four tasks was 4s per image.	AFW(AP=0.975),FDDB(AP=0.893)，PASCAL(AP=0.9246)	AFLW	RCNN
dubbed ScaleFace(Face Detection through Scale-Friendly Deep Convolutional Networks)	√	较热门	7fps for 900*1300 input images	wider face(ap=0.764)	WIDER FACE	ResNet RPN fast-rcnn
Faceness Net(From Facial Parts Responses to Face Detection: A Deep Learning Approach)	√	热	At the expense of recall rate, the fast version of CascadeCNN achieves 14fps on CPU and 100fps on GPU for 640*480 VGA images	PASCAL(AP=92.11%) AFW(AP=97.20%)	CelebFaces dataset	CNN
S3FD	√	一般	36fps	在wider face数据集上的mAP{EASY:93.7%,Medium:92.4%,Hard:85.2%}	WIDER FACE	RPN,SSD
mtcnn	只有测试代码	热	16fps on a 2.60GHz CPU and 99fps on GPU (Nvidia Titan Black)	FDDB(0.9504) WiderFace(EASY:0.851,Medium:0.820,Hard:0.607)	wider FACE,FDDB,AFLW	多任务级联卷积神经网络