search

xiaolai-sqlai / mobilenetv3

mobilenetv3 with pytorch，provide pre-train model
MIT License
1.6k stars 340 forks source link

Add large version which is same as the structure described in paper. #27

Closed TKONIY closed 4 years ago

TKONIY commented 4 years ago

The network you wrote is different from the one describe in paper. So I rewrote some parts of the your mobilenetv3_large module according to the origin module described in paper.

  1. SE should be placed between dw and pw.
  2. You add an extra BN in the last few layers.

The reason why I create this pull request:

  1. Your code performs really well. The repo is so good that I'd like to contribute to it.
  2. Since I'm only an ungraduate 19-year-old student, I can't afford to pretrain this model with imagenet. But I think we all want to find out whether the official version performs better.
  3. I‘m new here and now trying to be more integrated into GitHub community.😀

I can also rewrite the small version if you want me in.😘