This repo implements PyramidBox with pytorch The paper PyramidBox is based on VGG16. But this repo is based on resnet50.
Here, I implements data-anchor-sampling, LFPN, max-in-out layer.
The result on WIDER FACE Val set:
| AP Easy | AP Medium | AP Hard |
|---|---|---|
| 95.3 | 94.3 | 89.0 |
You can download model from Baidu!
With the model and test.py, you can get the same result on WIDER FACE Val set!
Before start, you should put the model into a file named weights
cd PyramidBox
mkdir weights
*Python3.5
*Pytorch3.1
*OpenCV3
*Anaconda: environment.yaml is the environment file, using conda env create -f environment.yaml to create the envrionment I test in my pc.
You can input the video or image