Detect a single 'main' object for cropping

[osecluna]

Hi, I'd like to use grounded SAM to find the 'main' object in an image, to assist with automated cropping. i.e. to detect the foreground then generate a coordinates of single bounding box to use for my crop.

Is there a way to understand the detected objects returned by grounded SAM so that I can automate this process and make a best guess at the 'main' foreground objects? Or alternatively a combination of all detected objects to define the foreground?

A single foreground object can be detected with other models such as InSPyReNet, but I already have groundedSAM implemented so would like to use this if possible.

Thanks!