Is there huge difference between dose scores in training set and validation set?
How much is the difference for your model?
Wondering the general situation, Thanks
It seems model performs better in validation set， because training set has much more “noise cases”.
Posted by: LSL000UD @ May 28, 2020, 3:05 p.m.really?
My situation is the opposite of yours.
The training set result was perfect but the validation set result was awful.
I wondered if overfitting is common for others.
Maybe you should use more data augmentations， in my experiments， training more iterations brings higher score（maybe 100k iterations with batch size 2）
Posted by: LSL000UD @ May 28, 2020, 3:54 p.m.Thanks for your advice.
Your training iteration number is far more than mine.
I will add more data augmentation.
Hope that it will work