Darknet: Can someone help me understand if it's training correctly?

Created on 9 Dec 2018 · 2Comments · Source: pjreddie/darknet

I created my own dataset and started training. I'm not sure what i'm looking at and how to tell if it's training correctly.

screenshot

What does the following mean?:

Avg IOU
Class
Obj
No Obj
avg
rate
seconds
images

Source

daylanc

Most helpful comment

AFAIK, there is no objective answer for how to tell if it is training correctly; perhaps gather a set of training inputs, test and if they fail, use them to train some more.

two places where I found useful information on the training output:

https://timebutt.github.io/static/understanding-yolov2-training-output/

https://github.com/AlexeyAB/darknet#when-should-i-stop-training

theshawnz on 20 Dec 2018

👍2

All 2 comments

AFAIK, there is no objective answer for how to tell if it is training correctly; perhaps gather a set of training inputs, test and if they fail, use them to train some more.

two places where I found useful information on the training output:

https://timebutt.github.io/static/understanding-yolov2-training-output/

https://github.com/AlexeyAB/darknet#when-should-i-stop-training

theshawnz on 20 Dec 2018

👍2

+1 this is a valid concern and would help a great deal if addressed.