Training on multiple concepts

[Deepayan137]

Hello,

First off, thank you for your excellent work. I have a question regarding the training process of the model. It appears from the code that the model can only be trained on one concept at a time, as specified by the --sks-name parameter.

Could you clarify how the model is capable of recognizing multiple concepts simultaneously, as demonstrated in Table 3, right column, of your paper?

Thank you in advance for your assistance.

[arctanxarc]

You can refer to our work's code: https://github.com/arctanxarc/MC-LLaVA. We extend Yo'LLaVA to handle multi-concept scenarios.

[Deepayan137]

Thank you for getting back to me. It looks like a very cool work. :)