search

TUI-NICR / ESANet

ESANet: Efficient RGB-D Semantic Segmentation for Indoor Scene Analysis
Other
236 stars 51 forks source link

Train my own dataset without void class #45

Closed amokame closed 2 years ago

amokame commented 2 years ago

Hello, I want to train the model on my own dataset, which has no void class. I changed the code in dataset_base.py and train.py as following.

dataset_base.py: image

train.py: image

The training code starts to run, but the mIoU for each epoch is very small even after 50 epoches. It seems to go wrong after my changes. Could you give me some advice about how to make it right?

Thank you.

danielS91 commented 2 years ago

It looks like you are excluding class 0 (which is a valid class in your scenario) in your training loss function. You should double-check the confusion matrix as well.

amokame commented 2 years ago

Thanks for your advise! Instead of changing the code, changing label to make it have the void class seems easier. I tried and it worked.