How to train the visual tokenizer?

[insoo-best]

Hi,

Thank you for sharing the outstanding work!

I ask some questions about the visual tokenizer.

Could you tell me more details about training the visual tokenizer?

The visual tokenizer is jointly trained with the infinity model? or separately trained in advance and frozen during training the infinity model?

What dataset did you use for training the visual tokenizer?

Any special training strategy for the visual tokenizer?

Thank you.