The trade-off between accuracy and speed in face detection has been widely researched. As many single-stage detectors has achieved remarkable performance on both sides, two-stage methods lost its advantages. Several experiments on the obstacles for single-stage face detector are conducted and the findings are summarized in the Tech Report(https://github.com/IssacCyj/OneStageFaceDetectionBaseline/blob/master/techreport.pdf). This repo provides an easy-to-adapt face detection baseline with some tricky settings commented in the codes. Hopefully it could provide some basic ideas for the beginners and researchers.
PS. The code is adapted from https://github.com/amdegroot/ssd.pytorch, where you can find more detailed info about the implementation.
The code is not yet clean and I'm still working on it in my spare time. Please read the report first and share with me if anything's wrong.
Welcome discussing.