I use train.py to train on my relabel ade20k dataset (150 to 5 classes) but couldn't get good result. mIoU is always at maximum 57%.
My training parameters:
Batch Size: 32
After plotting out the train loss and val loss graph, I found that the loss pattern is suspicious, it is shown below. Model is overfitting regularly after every epoch and I assume that's the reason why my model can't hit a higher mIoU. I have no idea why.
I use
train.py
to train on my relabel ade20k dataset (150 to 5 classes) but couldn't get good result. mIoU is always at maximum 57%.My training parameters:
After plotting out the train loss and val loss graph, I found that the loss pattern is suspicious, it is shown below. Model is overfitting regularly after every epoch and I assume that's the reason why my model can't hit a higher mIoU. I have no idea why.![loss](https://user-images.githubusercontent.com/21157726/69217705-3339e700-0baa-11ea-9770-a301299041dd.png)