Towards Comprehensive Scene Understanding: Integrating First and Third-Person Views for LVLMs

TL;DR

Large vision-language models (LVLMs) are increasingly deployed in interactive applications such as virtual and augmented reality, where a first-person (egocentric) view captured by head-mounted camera…

Opening excerpt from the authors’ abstract. source