pjreddie / darknet

Convolutional Neural Networks
http://pjreddie.com/darknet/
Other
25.7k stars 21.33k forks source link

Training coco data sets do not converge #1083

Open gllspeed opened 6 years ago

gllspeed commented 6 years ago

I am training the coco dataset according to the steps on the yolov3 official website. When iterating to about 90,000 generations, the loss will fluctuate around 5, no longer falling. Is this normal?

mercred commented 6 years ago

Experiencing similar issue, 40k iterations, avg.loss is about 4.5

jingtianyilong commented 6 years ago

I modified tinyyolov3 and ends up at about 6.5 with 50k iters and 6 with 100k iters on COCO