Closed vincentme closed 4 months ago
From the code, the current approach seems only test-time adapt based on query image of mask of one class.
Yes.
So if my application is to few-shot segment multiple class at once. Is this doable here? Is this doable here?
That's a valid question which I also think is worth addressing, but since previous work did only binary segmentation, I kept the single-class segmentation setting for this work.
One option to realize multi-class segmentation: For n classes of interest, split your support GT in n binary support masks and then do n subsequent predictions. Keep input query and support image the same and only change the support mask in each prediction.
Got it, thanks!
Hi
I have a question about the usgae. From the code, the current approach seems only test-time adapt based on query image of mask of one class. So if my application is to few-shot segment multiple class at once. Is this doable here?
Thanks!