Are labels given by catalog reliable?

[jacob975]

The true label is given by the method described in Harvey et al. (2007).
I feel like that is a simple mathematical model, based on the morphology of the data distribution on color-color diagram.
I am not sure it is reliable or not.

[jacob975]

What AI do during training process is catching-up the predictions that catalog provide.
This is no way to judge the labels with only one catalog.
We are going to train our AI with another catalog, e.g. H&L 2013.