search

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.66k stars 7.96k forks source link

Current Avg Loss #596

Open abeyang00 opened 6 years ago

abeyang00 commented 6 years ago

if the current avg loss says 0.222 during training, does it mean its loss is 22% or .22%?

And my loss isnt going down from .22 even after 15000 iterations, should i add more negative samples and stuff you mentioned to improve accuracy?

Thanks

AlexeyAB commented 6 years ago

Avg loss is measured as absolute value, it can't be %. Avg loss depends on model and number of iterations. You should not be guided by any particular meaning of avg loss, just wait until the avg loss will not stop falling.

A much more important indicator is the accuracy - mAP: https://github.com/AlexeyAB/darknet#when-should-i-stop-training