RetinaFace(RetinaFace:The Single-Stage Dense Face Location in the Wild,2019年出版)在TensorFlow 2.0中重新实现,提供预先训练的权重。重新设置50主干。
从retinaface导入RetinaFace导入CV2检测器=RetinaFace(";./data/retinafaceweight ts.npy";,false,0.4)img=CV2.。Imread(";./sample-images/WC_FR.jpeg";)Faces,Landmark=检测器。检测(img,0.9)。
这项工作在很大程度上是基于令人惊叹的洞察力团队使用Widderface-评估repo所做的Wide Face评估的原始实现,如果您使用此repo,请参考原始工作:
@inProcestions{Deng2020CVPR,title={RetinaFace:Single-shot Multi-Level Face in the Wild},作者={邓,建康和郭,贾和维韦拉斯,Evangelos和Kotsia,Irene和Zafeiriou,Stefan os},booktitle={CVPR},Year={2020}}