Problem with num_classes in instance segmentation

[agnieszkalysak]

Hi, I would like to ask another question:)

Doing the evaluation on ADE20k instance dataset, the AssertionError inside the script dinov.py appears, where there is no equality between self.sem_seg_head.num_classes == scores.shape[-1].

As I understand, the instance segmentation dataset will not have specific classes count - it is dependent of the instances count - in Detectron2 data each instance simply has unique intensity value, which is its "unique ID".

Could you please correct me, if I'm wrong, but this seems to be the blocker to see the results on ADE20k instance segmentation data. Thanks!