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

Warmup scheduler seems not working?

feiyuhuahuo opened this issue · comments

commented

Hi
In the supplementary material in your paper, you used warmup in the first 500 iterations. But I found that the learning rate keeps the same when training.
image
The warmup_method in your config is 'constant', don't you think it should be 'linear'?

Hi, the constant warmup assigns a smaller LR (1/3 of the original LR) for the first 500 iterations. I think the linear warmup should also be working well. I don't think the choice between constant and linear warmup will make a big difference here.