Resize closer to trained size for optimal accuracy - Githubissues

biubug6 / Pytorch_Retinaface

Retinaface get 80.99% in widerface hard val using mobilenet0.25.

MIT License

2.63k stars 774 forks source link

Resize closer to trained size for optimal accuracy #27

Closed xsacha closed 5 years ago

biubug6 commented 5 years ago

Thank you for your idea! I would merge it.