Home

Awesome

yolo-face-with-landmark

实现的功能

Installation

Clone and install
  1. git clone https://github.com/ouyanghuiyu/yolo-face-with-landmark
  2. 使用src/retinaface2yololandmark.py脚本将retinaface的标记文件转为yolo的格式使用,
  3. 使用src/create_train.py 创建训练样本

训练

python train.py --net mbv3_large_75 --backbone_weights \
./pretrained/mobilenetv3-large-0.75-9632d2a8.pth --batch-size 16 

测试

python evaluation_on_widerface.py
cd widerface_evaluate
python evaluation.py

demo

python demo.py

精度

Widerface测试

方法EasyMediumHardFlops
Retinaface-Mobilenet-0.25(Mxnet)0.7450.5530.232
mbv3large_1.0_yolov3(our)0.8610.7810.387405M
mbv3large_1.0_yolov3_light(our)0.8560.7700.370311M
mbv3large_0.75_yolov3(our)0.8530.7780.382334M
mbv3large_0.75_yolov3_light(our)0.8450.7660.365240M
mbv3samll_1.0_yolov3(our)0.7980.6960.3185M
mbv3small_1.0_yolov3_light(our)0.7590.6620.30091M
mbv3samll_0.75_yolov3(our)0.7680.6730.305174M
mbv3small_0.75_yolov3_light(our)0.7540.6470.29180M
方法EasyMediumHard
Retinaface-Mobilenet-0.25(mxnet)0.8790.8070.481
mbv3large_1.0_yolov3(our)0.9000.8820.707
mbv3large_1.0_yolov3_light(our)0.9000.8740.683
mbv3large_0.75_yolov3(our)0.8860.8710.694
mbv3large_0.75_yolov3_light(our)0.8810.8620.678
mbv3samll_1.0_yolov3(our)0.8560.8270.602
mbv3small_1.0_yolov3_light(our)0.8470.8070.578
mbv3samll_0.75_yolov3(our)0.8410.8150.584
mbv3small_0.75_yolov3_light(our)0.8320.7960.553

ps: 测试的时候,长边为320 或者 640 ,图像等比例缩放

测试

<p align="center"><img src="test_imgs/selfie.jpg"\></p>

References