SAM bbox auto-segmentation

[TddKim0712]

Using SAM with clicking or dragging bbox is currently available.
I just wonder if SAM's bboxing can be automatic.
For example, after I finished annotation so there are many bboxes matching their masks
so the automatic SAM will re-masking (rectangle -> polygon fitted in every objects)
as if SAM in CVAT now is able to handle bboxing.

Is there already a way to automizing SAM masking right now that i don't know?
Plz give me some idea that can make this possible.

[bsekachev]

Good approach for this purpose is to implement new annotations action. It will take all the bounding boxes (or filtered) and run SAM interactor function for each one. But it will require writing a new code, thus you need some skills in typescript and knowledge of CVAT code in some extent.

Example action doing tracking with OpenCV: https://github.com/cvat-ai/cvat/blob/develop/cvat-ui/src/utils/opencv-wrapper/annotations-actions/tracker-mil.ts