Difference between Classfication.MulticlassJaccardIndex (weighted = "macro") and segmentation.MeanIoU

[DeShinHwa]

Hi community, may I ask what is the difference between Classfication.MulticlassJaccardIndex(weighted="macro") and segmentation.MeanIoU. Do both actually interchangeable?

[Borda]

The main difference is their typical use case and input format. Classification.MulticlassJaccardIndex (weighted=“macro”) computes class-wise Jaccard index averaged over classes, mainly for classification tasks with discrete labels. segmentation.MeanIoU is designed for pixel-wise segmentation masks and averages IoU across classes.
Both measure intersection-over-union but differ in expected inputs and context, so they aren’t fully interchangeable. Use the one that matches your task’s data format and prediction type.