Home > other >  Darknet yolov3 training problems, the loss is not convergence, a lot - nan
Darknet yolov3 training problems, the loss is not convergence, a lot - nan

Time:09-23

Training is such result, framework downloaded from making darknet yolo, had a cat and dog the category of the detection, then tests in camera, all over the screen box, please solve!

CodePudding user response:

See you have a good training results, and I have the same problems, configuration files, CFG to change

CodePudding user response:

The landlord to solve a cold, I also am this problem, nan and box out a pile of more

CodePudding user response:

You can try to change the loss function, in a way should be able to solve the problem of loss convergence

CodePudding user response:

CodePudding user response:


The solution is what ao agghh ahhh

CodePudding user response:

Interface issues, you raise the threshold of above 0.7 for display, feel your threshold should be 0.1 or lower
  • Related