Hi @behxyz , in this MR, I remove the part of compute evaluation mask and ITC, since they are only focusing on the challenge based data format, I'm not sure if they should be added into MONAI. Could you please help to review this MR and if needed, let's discuss about the next step for the implementation. Thanks!

Hi @Nic-Ma @wyli , I met an error like this:

    if torch.is_tensor(probs):
        probs = probs.detach().cpu().numpy()

The above part got the mypy error:
monai/metrics/froc.py:52:17: error: Item "ndarray" of "Union[ndarray, Tensor]" has no attribute "detach" [union-attr]
However, it used mypy 0.8, which is the same as my local environment, and I did not get this error locally. Could you please help to take a look at this issue? Thanks!

The code is in here.

Hi @Nic-Ma @wyli , I met an error like this:

    if torch.is_tensor(probs):
        probs = probs.detach().cpu().numpy()

The above part got the mypy error:
monai/metrics/froc.py:52:17: error: Item "ndarray" of "Union[ndarray, Tensor]" has no attribute "detach" [union-attr]
However, it used mypy 0.8, which is the same as my local environment, and I did not get this error locally. Could you please help to take a look at this issue? Thanks!

The code is in here.

2 points:

1. you can try to use isinstance(probs, torch.Tensor) instead of torch.is_tensor() and try to submit PR again.
2. The reason why your local test is OK: your numpy is old, if you update to the latest numpy, you will get the error in mypy.

Thanks.

Hi @Nic-Ma @wyli , I met an error like this:

    if torch.is_tensor(probs):
        probs = probs.detach().cpu().numpy()

The above part got the mypy error:
monai/metrics/froc.py:52:17: error: Item "ndarray" of "Union[ndarray, Tensor]" has no attribute "detach" [union-attr]
However, it used mypy 0.8, which is the same as my local environment, and I did not get this error locally. Could you please help to take a look at this issue? Thanks!
The code is in here.

2 points:

1. you can try to use isinstance(probs, torch.Tensor) instead of torch.is_tensor() and try to submit PR again.
2. The reason why your local test is OK: your numpy is old, if you update to the latest numpy, you will get the error in mypy.

Thanks.

Thanks Nic,

My running environment is the latest MONAI docker, and the numpy version is 1.19.1, I upgraded it into 1.19.5 but still cannot get the same error.

the latest numpy 1.20 doesn't support python 3.6 which is the default python of the docker image...