How to get the visualization results of only instance masks without the original image?

How to get the visualization results of only instance masks without the original image?

When I used demo to save the images of inference, it usually shows the instance segmentation results on original images.

I want to ask how to get only instance segmentation results without original images.

Thanks!

Start by reading the demo's inference flow and the part that saves visualization images. Determine how the current output combines instance-segmentation results with the original image. Done means documenting a supported way to save only the instance masks, or clarifying that the demo does not provide that output.

Written by the indexing model from the issue text.