Strange metrics for panoptic deeplab output for custom dataset

[anushapv2000]

I am getting following inference output after training for 20k iterations for panoptic deeplab in detectron 2 framework .

The value of PQ and SQ is coming more than 100.

[07/11 05:52:50 d2.evaluation.panoptic_evaluation]: Panoptic Evaluation Results:

I registered my custom dataset in train_net.py file using
register_coco_panoptic(data_name, metadata, image_root, panoptic_root, panoptic_json, instances_json)
I have arranged things (42 classes) and stuff (38 classes) with proper ids.
@bowenc0221 ,can this sort of discrepencies arise due to wrong registration of data or something else?