Question regarding custom_imports and custom data pipeline class

[ivansabolic]

Cheers,

I have created a new pipeline class following this tutorial.
I am having trouble importing it in my pipeline. I have installed mmdet through pip.

Should I save my custom class in mmdet.datasets.pipelines package (part of source code) or am I able to add it to custom python package in my "project"? If former is the case, do I have to rebuild mmdet from the source code?

Latter would be preferable in my case, however I am having trouble adding it as part of custom_imports.
My class CustomPipeline is saved in my_file.py which is part of my_package situated on the same level as my train script.
Inside config I am trying to do
custom_imports = dict(imports=["my_package"], allow_failed_imports=False) and then inside the pipeline dict(type='CustomPipeline', ...)
but I am not able to import it. I have also tried adding from my_file import CustomPipeline inside my_package/__init__.py.

Any advice is appreciated!