Giters

kkhoot / PAA

A PyTorch implementation of the paper `Probabilistic Anchor Assignment with IoU Prediction for Object Detection` ECCV 2020 (https://arxiv.org/abs/2007.08103)

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Performance

qianwangn opened this issue · comments

commented

I implement PAA based on detectron2.
But only got 40AP without score voting, with 1x setting.
does any reasons that cost 0.5 points lower?

commented

Hi, thanks for your work! I don't know the exact reason because I didn't see your code, but the reason might be in the separation boundary. In the paper we reported that All boundaries (Fig. 3 (a) ~ (b)) show similar performance, but after repetitive experiments we found that Fig. 3 (c) is the most stable one. So I recommend you to try Fig. 3 (c). We will update our paper accordingly. Thanks again for your work!

commented

Closing as the paper has been updated.