0

I used a CNN model for segmentation task. the output training is like below :

enter image description here

how to know when should I stop training based on Loss and IoU?

enter image description here

enter image description here

Somaye K.
  • 13
  • 5

1 Answers1

0

Stop training when the loss is at minimum. When loss begins to increase(validation error is at its minimum) or accuracy begins to decrease, stop the training. In your image this appears to be Epoch 23, iteration 12600. IOU>0.5 is considered a good prediction.

B R
  • 36
  • 3
  • Thank you. I plot the average of loss for each epoch. I added the plot in the question. please look at it. – Somaye K. Dec 06 '20 at 08:18
  • I cannot tell if that is a plot of training loss or validation loss, but one method is to stop training at the point when performance on a validation dataset starts to degrade. This approach is called early stopping. – B R Dec 06 '20 at 08:41
  • it is training loss plot. I dont have validation loss. where can I find the validation loss based the first picture in the question? – Somaye K. Dec 06 '20 at 08:47
  • https://stackoverflow.com/questions/53232969/how-to-monitor-validation-loss-in-the-training-of-estimators-in-tensorflow – B R Dec 06 '20 at 09:01
  • @SomayeK. please accept my answer/ upvote if my answer was helpful! Thanks :) – B R Dec 06 '20 at 19:22