[Docs] What's wrong with the initialize function in classes

[sceddd]

📚 The doc issue

I have no idea how to create a custom datasets in mmengine with init method, i tried to create a fewshot dataset like this
Error

@DATASETS.register_module()
class FewShotDatasetLVISV1Dataset(BatchShapePolicyDataset, LVISDataset):

    METAINFO = {
        'classes': ('object',),
        'palette': [(220, 20, 60)]
    }

    def __init__(self, num_shot=5, repeat=False, *args, **kwargs):
        super().__init__(*args, **kwargs)
        print('FewShotDatasetLVISV1Dataset')
        print(num_shot)

        print("self num",self.num_shot)
        self.repeat = repeat

Correct

@DATASETS.register_module()
class FewShotDatasetLVISV1Dataset(BatchShapePolicyDataset, LVISDataset):

    METAINFO = {
        'classes': ('object',),
        'palette': [(220, 20, 60)]
    }

    def __init__(self, num_shot=5, repeat=False, *args, **kwargs):
        print('FewShotDatasetLVISV1Dataset')
        print(num_shot)
        print("self num",self.num_shot)
        self.repeat = repeat
        super().__init__(*args, **kwargs)

why it only run the super().init(*args, **kwargs) but don't continue with the rest of initialize block? And I realize that I have to implement the variables first before run the super class implementation, why the logic is so confuse? and no docs talking about it. I think you should create one for this annoying error.

Suggest a potential alternative/fix

No response