MeanIoU metric gives ValueError

[Barthas255]

Hi, Im testing DeepLabV3+ for 2 class segmentation task and wanted to change the metric. Accuracy is pretty useless in my scenario, since I'm trying to detect buldings on orthophotomap, and the rest is just background.
But when I try to change metric to IoU (or MeanIoU) I'm welcomed with this ValueError:

I tried looking for the answer on the internet, wasn't successful.
I tried to implement my own IoU like this:

but this sometimes gives values outside of range (0,1), which is bad.
Also I believe somehow y_pred and y_true don't have the same dimensions, which is odd for me:

I might be missing something obvious
Any solutions?

[DoraPist]

Hello everyone! I have the same issue... I would like to use MeanIoU as a metric, but I get the above error too. If someone knows how to solve it, please, let an answer here. It would be very helpful.
Thank you in advance!