Class names in checkpoint's meta data

How to save the class names of my custom dataset in the checkpoint’s metadata?

I don’t see any option in train.py or from skimming the tutorial on config here.

[mzr1996]

Well, in the tools/train.py, we will automatically add the CLASSES attribute of the dataset to the checkpoint’s meta argument. Usually, you only need to add the classes in the dataset configuration. And custom the meta dict in the config file may cause some other important meta keys missing. Here is the code: https://github.com/open-mmlab/mmclassification/blob/dc35eb62b61ab17decedb93c4d83ec6da5f93eff/tools/train.py#L161-L167

[Ezra-Yu]

By the way, we are going to add some documentation about Hook in #499, which includes CheckPointHook, and it is under review. Maybe you can read that and give us some suggestions.