I don't quite see how this would help OCR at all? or am I misunderstanding what kind of OCR you're thinking of?

Deepseek-OCR uses SAM V1 as a component in its pipeline already. It also does layout detection.

That sounds like ludicrous overkill to me.

for document layout! did you have success understanding document layout using SAM