About Dynamic image size input.

Hi,
Thank you for releasing this excellent repository. I want the dynamic size of the input image to get the mask instead of converting to 1024*1024 every time for inference. I have tried but have yet to succeed.
Can you give me some advice?
Thanks.
Best.

Start by reading the repository's inference example notebooks and identify where the input image is converted to 1024×1024 before mask generation. Compare that path with the requested dynamic-size behavior and determine whether the model supports it. Done means inference can accept the requested input dimensions without that conversion, with the resulting masks still behaving correctly.

Written by the indexing model from the issue text.