It's so nice of you to share the code with us,it's a wonderful work!
I know about ohem which is to select hard examples and re-train them,but what's the use of warmup?
waiting for your apply!
That is one of my attempts to solving the problem that joint training with OHEM will deliver slightly lower mAP by using a smaller learning rate at the beginning. But I failed, so please just ignore it.
It's so nice of you to share the code with us,it's a wonderful work! I know about ohem which is to select hard examples and re-train them,but what's the use of warmup? waiting for your apply!