SRTube: Video-Language Pre-Training with Action-Centric Video Tube Features and Semantic Role Labeling | Read Paper on Bytez