Training with already searched policy.

dvlab-research / SA-AutoAug

Scale-aware Automatic Augmentation for Object Detection (CVPR 2021)

Other

196 stars 22 forks source link

Training with already searched policy. #1

Closed qianwangn closed 3 years ago

qianwangn commented 3 years ago

The 6x schedule is too long for training. If I want to search augmentation policy in the first time and use the searched policy after, How can I do that? @yanwei-li Thanks.

yukang2017 commented 3 years ago

Hi,

The search don't need that long. You can follow maskrcnn-benchmark/search.sh for search. And then set cfg.AUTOAUG.LIST as your searched policy. And then train your model as the instructions in README.md.

Regards, Chen Yukang