KeyError: 'Dice. front' when training the segementation task

[SZUHvern]

Thank you for sharing the code!

There is an error occur when I trainning the segementation task:
File ".../site-packages/mmcv/runner/hooks/evaluation.py", line 389, in evaluate
return eval_res[self.key_indicator]
KeyError: 'Dice.front'

After inspection, I found that in /mmseg/core/evaluation/metrics. py, the function 'total_area_to_metrics' has the following definition:

Allowed_ Metrics=['mIoU ',' mDice ',' mFscore ',' medDice ']
But there is no 'if' item for 'medDice':
For metric in metrics:
If metric=='mIoU ':
...
If metric=='mDice':
...
I guess is any code missing here? or can I directly use [If metric=='mDice' or metric=='medDice ': ] ?

[DopamineLcy]

Thank you, and the function returns in

MRM-pytorch/Siim_Segmentation/mmseg/core/evaluation/metrics.py

Line 310 in 3730ba4

return {'Dice': dice}

when metrics == ['medDice']
the function 'total_area_to_metrics' should not be executed

[SZUHvern]

if mmcv.is_list_of(results, np.ndarray) or mmcv.is_list_of(results, str):

Do you mean that the program will execute here? But the variable ‘results’ I get here is a list of tuple

[DopamineLcy]

Variable ‘results’ here should be a list of np.ndarray

[SZUHvern]

Thanks,
But how can I get the mentioned list [np.array]? The annotation of type here is consistent with mine
https://github.com/RL4M/MRM-pytorch/blob/3730ba49a678eda870284205178bfe25627aaed8/Siim_Segmentation/mmseg/datasets/custom.py#LL397C17-L397C17