search

YonghengSun1997 / MSCA-Net

code for our MSCA-Net
15 stars 4 forks source link

About the experiment #7

Open wanwan-na opened 7 months ago

wanwan-na commented 7 months ago

Hello, I would like to ask whether your experiment (mean ± standard deviation) is The test output result, for example, The ISIC mean dice: 0.9239; The ISIC dice std: 0.1149 As a result? Is it 0.9239+-0.1149? 1

YonghengSun1997 commented 7 months ago

Yes. For each fold from five folds, we calculate the mean result on test set of each fold. Then the reported mean and std is calculated on these five results.

wanwan-na commented 7 months ago

Would you like to ask if the dice value is greater than 1 when 0.9239+-0.1149 is added, is this also normal?

YonghengSun1997 commented 7 months ago

What do you mean by "the dice value is greater than 1 when 0.9239+-0.1149 is added"?

wanwan-na commented 7 months ago

If the final dice is 0.9239+-0.1149, 0.9239+0.1149 is greater than 1, but shouldn't the dice value be between 0 and 1?

YonghengSun1997 commented 7 months ago

I can't find the "0.9239+-0.1149" result. I guess "0.9239+-0.1149" may be " (92.39 + 0.1149) %? The std is calculated on five results from five fold (std is calculated on five scalars). The five fold results is rather stable, and the std of these five results is rather small.

wanwan-na commented 7 months ago

I averaged the dice value and std obtained from the cross-verification of 50% fold respectively to get the final result. Is it correct to write it in yellow at the end? 2

YonghengSun1997 commented 7 months ago

Yes, I reported mean and std of 5 folds, like your results in yellow.

YonghengSun1997 commented 7 months ago

So ”0.9239+-0.1149“ is your fold 1 results. I think it is normal, because some cases is hard to segmentation, and have low dices （such as below 0.65）, which causes big std. Actually, the ISIC challenge treat iou below 0.65 as 0 when evaluation.

wanwan-na commented 7 months ago

Ok, thank you very much for your answer