I have a question about semantic segmentation

[dlgmltjdit]

I'm very impressed with your research, and I have a question about the semantic segmentation part. In the final step, the highlighted image is delivered to the LLM to extract semantics. Is there a separate training process for semantic segmentation, or is the extraction of semantics solely handled by the LLM without additional training?

[yhyang-myron]

Hi, Thanks for your interest in our work!
There's no separate training for semantic information acquisition.
It can be obtained directly from VLM or other vision models (such as GLIP, CLIP and GroundingDINO).

[dlgmltjdit]

Thank you for your response :)