search

matejgrcic / Open-set-M2F

Open-set Semantic Segmentation Built atop Mask-level Recognition
MIT License
11 stars 2 forks source link

How to modify the function to calculate the AUROC #1

Closed zzzyzh closed 1 year ago

zzzyzh commented 1 year ago

Hi! Thank you for your excellent work at first! When I wish to apply this method directly to the cityscapes dataset, I get the following error: ValueError: multiclass format is not supported The shape of mask_pred is [100, 1024, 2048], and the shape of mask_cls_ is [100, 20], what should I do? Best regards.

matejgrcic commented 1 year ago

Hi,

AUROC requires binary ground truth. That is, a dataset with both inliers and outliers (e.g. fishyscapes, road anomaly, ...). Outliers should be assigned to class 1 and inliers to class 0. However, Cityscapes does not contain outlier pixels hence evaluating AUROC is not informative on that dataset.

Hope this helps.

zzzyzh commented 1 year ago

Thanks a lot!