search

experiencor / keras-yolo2

Easy training on custom dataset. Various backends (MobileNet and SqueezeNet) supported. A YOLO demo to detect raccoon run entirely in brower is accessible at https://git.io/vF7vI (not on Windows).
MIT License
1.73k stars 784 forks source link

iou loss #255

Open ghost opened 6 years ago

ghost commented 6 years ago

is it ok if i change iou label to 1 instead of overlap of prediction and groundtruth because i think label need to be stable like 1, not unstable like overlap? thanks!

experiencor commented 6 years ago

@txdat Yes, it's possible. It's actually how YOLO3 works.