Reuse the Memory Bank for another video

I appreciate your wonderful work!

I have a question related to memory bank. In SAM2 paper, a memory bank includes two kinds of FIFO queue (recent frames, prompted frames), and corresponding object pointers based on mask decoder output token. Then if I store those memory bank elements at offline (after inference), can I use them to do inference on another video?

For example, **If I want to segment specific person in real-time setting (ex. webcam), I can give points prompt to a video which take that figures of the person and then store the memory bank of the video.** It's like training custom class in supervised learning.

Thank you for watching.