[feat] Is there a method similar to the Meta SAM demo that generates prompt points from parsed images?

Hi, thank you so much for the **amazing work** on this project! The design and implementation are really impressive, and the examples have been very helpful.

I have a question regarding automatic prompt generation.
In the Meta SAM demo, the system can parse an image and automatically produce prompt points or prompt regions for the segmentation model.

**Is there any method, module, or recommended approach in this project that can similarly generate prompt words/points based on image parsing results?**
If not, is there any suggested direction or component I could look into to implement this feature?

Thanks again for your great work and for open-sourcing this project!

**example：**