基于特征图融合的小尺寸人脸检测

Small Size Face Detection Based on Feature Map Fusion

人脸检测是指从输入图片或视频中找到人脸的精确位置并确定其大小。为了应对尺度多样性特别是小尺寸人脸给人脸检测任务带来的困难,文中提出一种新的基于特征图融合的小尺寸人脸检测方法。首先,合理地选择待检测特征图,使用不同的特征图检测不同大小的人脸。然后,通过将较深的特征图和较浅的特征图进行融合,合理地引入上下文信息,从而提高小尺寸人脸的检测精度。在NVIDIA GTX TATAN X上,使用WIDERFACE数据集对所提方法进行训练和测试,其在WIDERFACE 3个测试子集上的测试结果分别为88.9%(hard),93.5%(medium),94.3%(easy)AP,检测速度为39 fps,其检测精度和检测速度均优于其他优秀的检测方法。

Face detection is finding and locating all faces from the input pictures or videos.In order to solve the difficulties caused by the diversity of face size,especially small-sized faces,a new single shot small-scale face detection method is presented based on feature map fusion.The method first selects the feature map to be detected reasonably,and uses different feature maps to detect faces of different sizes.Then,by combining the deep feature map and the shallow feature map,the context information is introduced reasonably,thereby improving the detection precision of the small-sized face.The proposed model is trained and tested on the NVIDIA GTX TATAN X using the WIDERFACE dataset.The results on the three test subsets of WIDERFACE are 88.9%(hard),93.5%(medium),94.3%(easy)AP,at 39 fps.It is superior to other excellent detection methods in both detection accuracy and detection speed.