Skip to content

Next Gen Face detection based on YOLOv7

License

Notifications You must be signed in to change notification settings

monoloxo/yolov7-face

 
 

Repository files navigation

yolov7-face

Method Test Size Easy Medium Hard Link
yolov7-tiny-leak 640 93.2 91.3 83.0 google
yolov5s 640 94.0 92.1 84.5 google
yolov7-lite-e 640 91.1 87.5 70.9 google

Demo

Download the model place into weights folder, then run:

python detect.py --weights weights/yolov7-tiny-face.pt --view-img

Deploy

事实上这个仓库主要是验证wnnx在复杂模型下的推理能力,当然这当中存在着非常多的坑,最终,我们可以做到NCNN里面做不到的东西,后处理几乎全部融合在模型里面,外部只需要做一个score判断即可。

wnnx的推理结果:

第一张图为yolov7-tiny,后面两张图为 yolov7-lite-e,单线程在CPU下跑的realtime,多线程更快,fp16更快,精度还可以。

Export

export to onnx and torchscript:

python export.py --weights weights/yolov7-tiny-face.pt

Dataset

WiderFace

yolov7-face-label

Test

Demo

References

About

Next Gen Face detection based on YOLOv7

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 97.8%
  • Shell 2.0%
  • Dockerfile 0.2%