Dreambooth for the inpainting model

[Nomination-NRB]

Why use a random mask instead of a region-specific mask when using Dreambooth for the inpainting model. For example, if my training picture is a dog, why do I use random mask instead of the mask corresponding to the dog? I see that the training input only has pictures, there is no corresponding mask diagram

[sayakpaul]

What's your reference from our codebase?

[Nomination-NRB]

What's your reference from our codebase?

I use this file /examples/research_projects/dreambooth_inpaint/train_dreambooth_inpaint.py, and I found that a function called random_mask, so I want to know if my training picture is a dog or anything else, why do I use random mask instead of the mask corresponding to the what I need to train

[sayakpaul]

why do I use random mask instead of the mask corresponding to the what I need to train

I think that is there to prevent overfitting. But I am not well aware of the script. So, I will suggest tagging the contributor who added the script.

[Nomination-NRB]

ok, thanks for your help