Can SAM v2 be prompted with an image + segmentation to find similar objects in another image?

[IslamAlam]

I’d like to know if it’s possible to use SAM v2 (or other segmentation models) in the following way:

• Provide an image and its segmentation (e.g., a point or polygon) as a prompt.
• Use that segmentation to guide the model to find and segment similar objects in another target image.

The goal is to maintain consistent labeling across images—for example, if I label a few buildings in one image, I’d like the model to segment all buildings in another image with the same target value.

Questions:

1. Is this possible to achieve directly with SAM v2?
2. If not, what’s the best approach (or model pipeline) to accomplish this kind of class-consistent segmentation across images?

Thanks in advance for any guidance!

PS: @giswqs thank you so much for the great work that you have done and the efforts into this package.