How to train Learning Bounding Box Adjusters?

Question

How to train Learning Bounding Box Adjusters?

HMCCMH opened this issue 3 years ago · 1 comments

Hi DongSky,
I have a question that how to train learning bounding box adjusters? You have give the pth files such as LBBA_final.pth for using train_wsod.sh. But if I want to train the LBBA model without the lbba_init.pth, what should I do?

Answer 1 · 2022-04-25T10:28:29.000Z

Hi HMCCMH
Just like the description in the main paper, to obtain LBBA_init.sh, you just need to conduct the standard supervised object detection training algorithm (e.g., Faster R-CNN) on the auxiliary dataset (e.g., COCO-60).