Model and Code for Accurate Foreground Object Cut-Out

I am trying to cut out all foreground objects from a given image using the **sam2.1_hiera_large.pt** model.
However, the results I achieve are not as accurate as those produced on the demo portal:
[Segment Anything Demo](https://segment-anything.com/demo).

Could you please provide details on:

1. The exact model being used for the demo on the portal.
2. The code or approach implemented to achieve the highly accurate foreground object cut-outs showcased in the demo.

Thank you for your assistance!