different segment result between vit_l and vit_h

[jack-gits]

In my picture, there's a mobile, I want to segment it out.
but when I use vit_l, the mobile can be segmented out correctly.
but using vit_h, the mobile can't be segmented out.
please refer to the codes in grounded_sam_demo.py

original image:

segment result by vit_l:

segment resulb by vit_h: