Enhance deepedit transforms to do label check

[yiheng-wang-nv]

Errors should be raises if input wrong label_names

cc @diazandr3s @Nic-Ma

[diazandr3s]

Errors should be raises if input wrong label_names

cc @diazandr3s @Nic-Ma

Hi @yiheng-wang-nv,

This is a good point. One thing we could do is to use .lower() for the label names.
However, a minimum typo in the label name would be recognised as a completely different label.
Maybe with regular expressions?

[yiheng-wang-nv]

Hi @diazandr3s , thanks for the suggestion. I think for the first version, we can simply add .lower() only. I will submit a PR soon.

[yiheng-wang-nv]

I rethink about this requirements, the current logic is like:

# Creating guidance for all clicks
all_guidances = {}
for key_label in self.label_names.keys():
clicks = d.get(key_label, [])
clicks = list(np.array(clicks).astype(int))
all_guidances[key_label] = self._apply(clicks, factor)
d[self.guidance] = all_guidances

in the transform level, it may not be able to check input data keys. For example, if the input data and label names are like:

data = {'image': image_id, 'spleen': [[1, 2, 3,], [1, 2, 2]], 'bounding_box': [[2, 3], [4, 5]}
label_names = {
        "spleen": 1,
        "background": 0
    }

We cannot consider bounding_box as an error, since it may be used for other purpose, such as for the next transform or for loss calculation, ...

HI @diazandr3s , what do you think? Maybe we can close this issue right now