replace your picture in '/picture/demo.jpg' and run 'python predict_m.py'

the result is '/picture/result_demo.jpg'

some of the predict code is copy and rewrite from:faceboxes

wider result:

easy	medium	hard
0.846	0.840	0.455

demo:

update @ 2018/11/25:

we add rotation augmentation and train a new model,these model run the same time but more robust to each rotation of the picture:

demo:

usage:

replace the checkpoint to /weight/msos_pnet_rotate.pt & /weight/msos_onet_rotate.pt