GroundSight: Augmenting Vision-Language Models with Grounding Information and De-hallucination