hi,I meet with a question that train's miou is far heighter than test's,can you give me some advice?
Gayou333 opened this issue · comments
In principle, this means overfitting. Maybe you can add more data to train your model, and you can train for less iterations. I do not think a large gap between the performance on training and test set is always a bad thing. If the performance on the test set is good, then we would still accept the large performance gap.
I am closing this since question is answered.