ValueError: Invalid segmentation mask

[eppane]

Hello,

Got the following error when using mmdet model (rtmdet-ins):

 res = get_prediction(
          ^^^^^^^^^^^^^^^
  File "/mmdet/lib/python3.11/site-packages/sahi/predict.py", line 100, in get_prediction
    detection_model.convert_original_predictions(
  File "/mmdet/lib/python3.11/site-packages/sahi/models/base.py", line 168, in convert_original_predictions
    self._create_object_prediction_list_from_original_predictions(
  File "/mmdet/lib/python3.11/site-packages/sahi/models/mmdet.py", line 305, in _create_object_prediction_list_from_original_predictions
    object_prediction = ObjectPrediction(
                        ^^^^^^^^^^^^^^^^^
  File "/mmdet/lib/python3.11/site-packages/sahi/prediction.py", line 79, in __init__
    super().__init__(
  File "/mmdet/lib/python3.11/site-packages/sahi/annotation.py", line 526, in __init__
    raise ValueError("Invalid segmentation mask.")

ValueError: Invalid segmentation mask.

The validity of (boolean) mask is already verified here by checking if bbox can be extracted from it, but when it gets converted to coco-formatted polygon, the bbox cannot be extracted anymore (i.e. this fails).