[Feature Request] Annotated Segment Output

[unusualbob]

Can we get an 'annotation' mode where the model is used simply to annotate what it thinks different parts of the image are? I ask because sometimes the model has incorrectly labeled something I want to keep but I can't seem to identify what it thinks that object is so I can't get it to keep that part.

For example, a part of someones shirt was being ignored despite the shirt itself being correctly included in the prompt. The shirt had patterns so I wasn't sure if the mask was less than ideal, or if it thought something was in front of the shirt.

There seems to already be an example of how to do this here, but I am not familiar enough with python to add it to your extension:
https://github.com/facebookresearch/segment-anything/blob/main/notebooks/automatic_mask_generator_example.ipynb

Only needed addition to that would need to be to show what the clip labels are. If you wanted to get really fancy you might be able to make it so you could just click each segment you want to keep.

I suspect this will help a lot with people who are having trouble selecting things using the input prompt.