arXiv · AI· Abhishek Tyagi, Tanmay Jain, Vilas Biradar, Krithi Krishna Koduri, Misba Hul Bashir Masoodi, Swetha Rapolu, Hardik Rughwani, Rachana Kotha, Anjan Kumar Kaipa, Anudeep Katrevula, Rakesh Garlapati, Aniruddha Pratap Singh, Rajendra Patel, Abhay Tyagi, Sundeep Lakhtakia, Mohan Ramchandani, Rakesh Kalapala, D. Nageshwar Reddy·· 10 天前AI 评分35
EndoSCOP-V:用于内镜报告与临床推理的多模态模型多轮视频理解评测框架
EndoSCOP-V: A Multi-Turn Video Understanding Evaluation Framework for Multimodal Models in Endoscopy Reporting and Clinical Reasoning
AI 导读
EndoSCOP-V 是一个基于视频的评测框架,包含 1,307 段内镜视频和 20,274 个问答轮次,由 1,033 份临床医生审核的模板确定性生成。 该框架采用马尔可夫上下文剪枝协议逐轮评估,15 名认证内镜医生组成的专家组确立了 78% 的严格准确率基准。 最佳的 27B 级开源模型严格准确率仅达 59.8%,在量化测量和长时程视频干预推理方面仍存在显著差距。
来源:arXiv · AI · arxiv.org