Title:	Towards Reliable Large Vision-Language Models
DNr:	Berzelius-2024-249
Project Type:	LiU Berzelius
Principal Investigator:	Xixi Liu <xixil@chalmers.se>
Affiliation:	Chalmers tekniska högskola
Duration:	2024-07-21 – 2025-02-01
Classification:	10207
Keywords:

Abstract

Large Vision-Language Models (LVLMs) are extensively utilized as foundational models in numerous computer vision tasks. However, a notorious problem with LVLMs is their tendency to hallucinate, meaning they often identify objects that are not present in the image. This project aims to mitigate these hallucinations and enhance the reliability of LVLMs.