Text this: Enabling High-Level Worker-Centric Semantic Understanding of Onsite Images Using Visual Language Models with Attention Mechanism and Beam Search Strategy.