why do we need change post_nms_topN?

Open 978326187 opened this issue 5 years ago • 2 comments

https://github.com/fanq15/FSOD-code/blob/2e534649b57791182b6c93153d279ecfa554ecb0/lib/modeling/collect_and_distribute_fpn_rpn_proposals.py#L74 why do we multiply a scale to change post_nms_topN in this line? and coco's background is more complicate than imagenet. is this a little unfair using fsod to compare with coco, such that coco is more difficult to train?

Jun 17 '20 05:06 978326187

I can not open this link. I use the original code in this part.What do you mean the comparison between fsod and coco?

Jun 27 '20 05:06 fanq15

post_nms_topN = int(cfg[cfg_key].RPN_POST_NMS_TOP_N * cfg.FPN.RPN_COLLECT_SCALE + 0.5)this line. and the comparison is the following pic.

Jun 27 '20 06:06 978326187