Unexpected behavior with zero width or zero height bool mask

[vincent-vdb]

Hi and thanks a lot for your work!

When running the library, we encountered an issue in annotation.py, when the output bool mask of get_bbox_from_bool_mask() is None. It happened for example because of having a zero width or zero height mask, thus raising a ValueError.

[fcakyon]

Thanks a lot for reporting your issue @vincent-vdb!

What type of instance segmentation model are you using?

[vincent-vdb]

Thanks for your quick answer. We are using MMDet, we actually have a simple fix proposal, but it seems I can't push a new branch to propose it.

Basically, in annotation.py, line 539, switching the raise ValueError to self.mask = None fixes the issue on our side.

[fcakyon]

@vincent-vdb you can fork this repo, push your update to a branch in your fork, then open a PR to this repo 🚀. We are open to PRs 👍

I have opened a PR to fix this can you test it by installing the sahi from branch handle-invalid-mask?

You have to first uninstall your current sahi installation by:

pip uninstall sahi -y

Then install sahi from branch:

pip install git+https://github.com/obss/sahi.git@handle-invalid-mask

If you confirm this fix we can merge it 👍

[vincent-vdb]

This PR does not fix the issue on my side :/

Here is the traceback if it can be of any help:

Traceback (most recent call last):
  File "/opt/conda/bin/sahi", line 8, in <module>
    sys.exit(app())
  File "/opt/conda/lib/python3.8/site-packages/sahi/cli.py", line 24, in app
    fire.Fire(sahi_app)
  File "/opt/conda/lib/python3.8/site-packages/fire/core.py", line 141, in Fire
    component_trace = _Fire(component, args, parsed_flag_args, context, name)
  File "/opt/conda/lib/python3.8/site-packages/fire/core.py", line 466, in _Fire
    component, remaining_args = _CallAndUpdateTrace(
  File "/opt/conda/lib/python3.8/site-packages/fire/core.py", line 681, in _CallAndUpdateTrace
    component = fn(*varargs, **kwargs)
  File "/opt/conda/lib/python3.8/site-packages/sahi/predict.py", line 477, in predict
    prediction_result = get_sliced_prediction(
  File "/opt/conda/lib/python3.8/site-packages/sahi/predict.py", line 247, in get_sliced_prediction
    prediction_result = get_prediction(
  File "/opt/conda/lib/python3.8/site-packages/sahi/predict.py", line 88, in get_prediction
    detection_model.convert_original_predictions(
  File "/opt/conda/lib/python3.8/site-packages/sahi/model.py", line 144, in convert_original_predictions
    self._create_object_prediction_list_from_original_predictions(
  File "/opt/conda/lib/python3.8/site-packages/sahi/model.py", line 344, in _create_object_prediction_list_from_original_predictions
    object_prediction = ObjectPrediction(
  File "/opt/conda/lib/python3.8/site-packages/sahi/prediction.py", line 75, in __init__
    super().__init__(
  File "/opt/conda/lib/python3.8/site-packages/sahi/annotation.py", line 539, in __init__
    raise ValueError("Invalid boolean mask.")
ValueError: Invalid boolean mask.

[fcakyon]

@vincent-vdb found the issue and updated just now. Can you please remove sahi and install again from branch and check if its fixed?

You have to first uninstall your current sahi installation by:

pip uninstall sahi -y

Then install sahi from branch:

pip install git+https://github.com/obss/sahi.git@handle-invalid-mask

[vincent-vdb]

It seems to be working now, thanks. At least I did not encounter the issue for the moment on the same data. Hope you can merge this PR soon!

[fcakyon]

@vincent-vdb it is merged, you can install from master till next release!

Thanks for your feedback 🚀