作者位次 #4/8ACL 2026Poster· Natural Language Processing
Putting Captions to the Test: Evaluating Video Caption Quality through Multiple-Choice Question Answering
Zizhen Wang · Bo Feng · Zhengfeng Lai · Shiyu Li · Yang Lu · Meng Cao · Ping Huang · Xiaoming Simon Wang
摘要
Evaluating video captioning remains a critical challenge for Visual Large Language Models (VLLMs). Existing metrics primarily rely on matching generated text against ground-truth references. This paradigm suffers from the “one-to-many” nature of video description, where high-quality captions are...