TypeError: 'NoneType' object is not callable

[IambecauseIthink]

I could run scripts/eval.py smoothly, but when running scripts/train.py, it went wrong with "TypeError: 'NoneType' object is not callable
", could you please help me to figure out?

[wangsen99]

Our model needs criterion (loss function) during training, you should simply modify the core/models/fdlnet_deeplab.py following the awesome-semantic-segmentation-pytorch" or modify the scripts/train.py .

[IambecauseIthink]

Thanks for your answer. I think the main difference in core/models/fdlnet_deeplab.py is:
if self.training: return self.criterion(outputs, gts) than other models in awesome-semantic-segmentation-pytorch
I tried to print self.criterion and self.model.criterion in scripts/train.py:
print(self.criterion) # The output is MixSoftmaxCrossEntropyLoss() print(self.model.criterion) #The output is None

I don't know how to modify it, and just simply removed this and the init of the criterion in the class, then it can start training. I'm not sure how this will affect, could you please help?

[wangsen99]

You can just remove the self.criterion and use MixSoftmaxCrossEntropyLoss() to train your own model, but our paper uses a multi-loss to gain better performance. You can try to implement it. ^_^

[IambecauseIthink]

Thanks a lot!