How to get AR@30 and AR@50?

[xishanhan]

I used train_net.py to train RPN model on coco2017 and got the model performance of AR@100 and AR@1000.
However, I want to get more evaluations of this RPN model, such as AR@30、 AR@50, so, how should I modify the code?

[xishanhan]

I have solved this problem by modifying /detectron2/evaluation/coco_evaluation.py#L315, from 'for limit in [100, 1000]' to 'for limit in [1, 10, 30, 50, 100, 1000]'