det_thresh in yaml files.
Opened this issue · 3 comments
cxmscb commented
Thanks for the sharing! I have a confusion about the setting of det_thresh in the superpoint_train.yaml
and magic_point_coco_train.yaml
files. Why is the det_thresh 0.001 rather than 0.015?
shaofengzeng commented
I think this is mainly for better performances
kaiyi98 commented
I think this is mainly for better performances
但是0.001的阈值会导致点特别多,做nms时间瓶颈很大,在训练superpoint的时候gpu利用率就很低,训练很慢,请问有什么方法解决的吗?我暂时是把这阈值调大了。
shaofengzeng commented
直观上,调小一些,会有保留更多的点参与训练,能保证效果。但,这些都不是确定的~,您可以尝试其它值试试,看效果如何