yolov7-face

New feature

• Dynamic keypoints
• WingLoss
• Efficient backbones
• EIOU and SIOU

Method	Test Size	Easy	Medium	Hard	FLOPs (B) @640	Google	Baidu
yolov7-lite-t	640	88.7	85.2	71.5	0.8	google	gsmn
yolov7-lite-s	640	92.7	89.9	78.5	3.0	google	3sp4
yolov7-tiny	640	94.7	92.6	82.1	13.2	google	aujs
yolov7s	640	94.8	93.1	85.2	16.8	google	w72z
yolov7	640	96.9	95.5	88.0	103.4	google	jrj6
yolov7+TTA	640	97.2	95.8	87.7	103.4	google	jrj6
yolov7-w6	960	96.4	95.0	88.3	89.0	google	-
yolov7-w6+TTA	1280	96.9	95.8	90.4	89.0	google	-

Dataset

WiderFace

yolov7-face-label

Test

Demo

• ncnn_Android_face

• yolov7-detect-face-onnxrun-cpp-py

References

• https://github.com/deepcam-cn/yolov5-face

• https://github.com/WongKinYiu/yolov7

• https://github.com/TexasInstruments/edgeai-yolov5/tree/yolo-pose

• https://github.com/ppogg/YOLOv5-Lite

保存特征

python detect.py --face_lib 人脸库图片目录 --feat-save-dir 人脸特征保存的目录, 
--img_size  图片尺寸, --conf_thres 0.7, --save_crop, --save_feature, 
--weights 检测模型pt, --rec_model 识别模型onnx

识别过程

python detect.py --source 输入源, --img-size 图片尺寸, --conf-thres 0.5, 
--feat_read_dir 人脸特征读取目录 --mode adaface --weights 检测模型pt, 
--rec_model 识别模型onnx