biubug6 / Pytorch_Retinaface

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

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

How to use the weights "mobilenet0.25_Final.pth"?

grapefruitL opened this issue · comments

I use following cmd to do evaluation

python test_widerface.py --trained_model mobilenet0.25_Final.pth --network mobile0.25

but it turns out that "mobilenetV1X0.25_pretrain.tar" is used(here) and it is the imagenet pretrain weights, which is uesed for training and can not be used to detect face

simply replacing "mobilenetV1X0.25_pretrain.tar" with "mobilenet0.25_Final.pth" didn't work
can anyone using author's mobilenet pretrain weights successfully help me?

mobilenetV1X0.25_pretrain.tar is for pretrain, you can see the data/config.py/cfg_mnet dict, if you just want inference, you can set pretrain False.