error in ProbOhemCrossEntropy2d

Reagan1311 / DABNet

Depth-wise Asymmetric Bottleneck for Real-time Semantic Segmentation (BMVC2019)

https://github.com/Reagan1311/DABNet

MIT License

140 stars 36 forks source link

Closed Serge-weihao closed 5 years ago

Serge-weihao commented 5 years ago

I got this error in ProbOhemCrossEntropy2d, when trainning on cityscapes. 屏幕截图(2)_LI

Reagan1311 commented 5 years ago

May I ask what code did you change? It seems that you didn't set the ignore label in the loss function.

Serge-weihao commented 5 years ago

May I ask what code did you change? It seems that you didn't set the ignore label in the loss function.

ignore label is setted in your train.py, which is 255, and the problem seems to be caused by tensor type.

Serge-weihao commented 5 years ago

I replace 1-valid_mask with ~valid_mask, and the code for trainning cityscapes can run now.

Serge-weihao commented 5 years ago

And I want to know the mIoU of 69.57% is from val set or test set?

Reagan1311 commented 5 years ago

This result were trained on the training set and tested on the val set.