config for conditional training?

[entangledothers]

trafficstars

I'm trying to figure out how to train a conditional model, but am a little stumped as to the correct way to setup the config, my data structure looks this:

root_folder -- category_01_folder -- category_02_folder -- category_03_folder ... etc.

My question is how do I format the config, say from the custom_vqgan config file to train this as a conditional model, and additionally, the text files we should generate for train and test, how are they formatted in this regard?

Thanks in advance for any help!

[areiner-novetta]

I came here looking for this as well. I found this config from the conditional image net (CIN) model linked in the README, which looks like this:

model:
  ...
  params:
    first_stage_key: image
    cond_stage_key: class_label
    ...
    cond_stage_config:
      target: taming.modules.util.Labelator
      params:
        n_classes: 1000

However, it's not clear how the labels for each image are supposed to be supplied during training.