When I use the demo demo on huggingface for image segmentation, I get great results:
But when I use the published pre-trained model for segmentation, I get poor results:
Do you know what the problem is? What can I do to reproduce the results of the demo?