why mobilev1/v2 only use middle features to predict?

Tencent / ObjectDetection-OneStageDet

单阶段通用目标检测器

Other

1.97k stars 499 forks source link

Open wangxinzhe0617 opened 5 years ago

wangxinzhe0617 commented 5 years ago

why mobilev1/v2 only use middle features to predict? The result is better?

mileistone commented 5 years ago

I just make a demo, you can modify it.

wangxinzhe0617 commented 5 years ago

ok，Thank you for your reply.

hilmandayo commented 5 years ago

I also want to know how to modify it using all features...