[Feat]: Generate captions on masked images

[zlstringham]

Describe your use-case.

I can generate image masks for masked training. However, I would also like to generate my captions based on the resulting masked images.

For example, using WD14 tagging, it makes little sense to tag all of the things that are not part of the masked image.

What would you like to see as a solution?

Because masked training is optional even if masks exists, this should be an optional toggle in the "Batch generate captions" dialog. Something to the effect of "Masked captioning" [true/false]

Have you considered alternatives? List them here.

I suppose I could write a python script to combine the images and masks and create a new dataset based on that, but it seems unnecessary to require this.