Closed Chi-Zaozao closed 4 years ago
That means that your detection predictions generate nonreasonable corner values. The accuracy are really low + out of any possible range. e.g. 9.67536e+12 9.43265e+13 9.67536e+12 9.43265e+13 -9.67536e+12 -9.43265e+13 -9.67536e+12 -9.43265e+13
Thank you very much for your reply.
But why is the program stuck here?Do you know how to avoid this?
And It only happens at the val stage.
This is because your model doesn't train well. If that is the log of your model this, the loss is a few thousand, which means that the model doesn't predict anything useful... You may need to tune the learning rate, and also it seems there are only 9* batch_size examples in the dataset which is probably too small. Additionally, you need to check that your data format / annotations are correct. First try to get the loss done to some one digit number and then check other stuff
When I train second model on my own dataset, strange output occurs at val stage as following:
I never add a output like this, and I don't know when will it stop. I will be appreciate if you could explain it. Thank you very much!