Validation pixAcc: 1.000, mIoU: 0.500 about awesome-semantic-segmentation-pytorch HOT 9 CLOSED

no problem!
I have solved the problem!

you are right, the problem is my label. I have 2 classes,but my labels are more than two classes!
thank you very much!

from awesome-semantic-segmentation-pytorch.

Maybe due to the dataset or the dataloader.

from awesome-semantic-segmentation-pytorch.

my data has two classer :car and background,and my lable's pix is 255.I change /core/data/dataloader/pascal_voc.py as my car.py. I add def get_psp_resnet50_car(**kwargs):
return get_psp('car', 'resnet50', **kwargs) in pspnet. no error occour but the the loss is not right.could you tell me what should I do next? thank you

from awesome-semantic-segmentation-pytorch.

The label value shoud be set to 0 and 1.

from awesome-semantic-segmentation-pytorch.

I haven't solved the problem.could you tell you if I want to train my own data,what should I do ?

from awesome-semantic-segmentation-pytorch.

You should ensure that the value is 1 for car and 0 for background in label.

from awesome-semantic-segmentation-pytorch.

Could you tell me where should I check my labels? thank you very much

from awesome-semantic-segmentation-pytorch.

Would u mind showing me your image sample and dataloader script？

from awesome-semantic-segmentation-pytorch.

Great!

from awesome-semantic-segmentation-pytorch.