A question about conditional generate

[holydick99]

In this paper team mention that use pseudo label to generate the corresponding data. So I'm a little confused about the process of training diffusion model and how use label to generate data? In my view, picture is the input(x) then label is the condition, is this model just need a label can generate a data? or need sth else?

[yyyouy]

I'm sorry to reply you so late. This is a scene that mainly involves label to image, so only one label is needed as a condition.