if one anchor is asigned to multi label, then should choose the max iou truth label

[ZouJiu1]

during training stage, if one anchor is asigned to multi label, then should choose the max iou truth label.
the default action is choose the last truth label, but in fact, one anchor should choose the truth label which has the max iou with the anchor.

[stephanecharette]

Note that this repo has been abandoned. The new Darknet/YOLO repo is at https://github.com/hank-ai/darknet

[ZouJiu1]

Ok, I will pull request to it

[ZouJiu1]

I fix some logical problam, now it can make and train normaly. I also commit it to that repository, if possible, please approve and write to this repository https://github.com/AlexeyAB/darknet too

[cenit]

this repo is not fully abandoned, it's just under very very low maintenance, and is still the only reference for any non-company-affiliated darknet versions ;)

[stephanecharette]

Alexey has stated he's no longer working on it, and the last real work on this repo was over 2 years ago.

This is why Hank.ai has freely donated resources and developers to carry on development of Darknet/YOLO.