arXivDaily arXiv每日学术速递 周一至周五更新
​通知:联邦节假日调休:哥伦布日 / 原住民日(Columbus Day / Indigenous Peoples' Day),今日(13日)无更新。
arXiv 2601.10228cs.CVcs.MMeess.IV

优化多模态大语言模型以实现视角视频理解:HD-EPIC VQA挑战的解决方案

Optimizing Multimodal LLMs for Egocentric Video Understanding: A Solution for the HD-EPIC VQA Challenge

Sicheng Yang, Yukai Huang, Shitong Sun, Weitong Cai, Jiankang Deng, Jifei Song, Zhensong Zhang

AI总结:

本文提出一种优化多模态大语言模型的方法,通过预处理、微调和时间链式思考提示,在HD-EPIC VQA挑战中实现了41.6%的准确率。

摘要:

多模态大语言模型(MLLMs)在复杂的视频问答基准测试如HD-EPIC VQA中面临挑战,原因包括模糊的查询/选项、较差的长时间域推理能力和非标准化的输出。我们提出一个框架,整合查询/选项预处理、领域特定的Qwen2.5-VL微调、一种新的时间链式思考(T-CoT)提示以进行多步推理,以及稳健的后处理。该系统在HD-EPIC VQA上实现了41.6%的准确率,突显了在 demanding 视频理解中需要整体流程优化的必要性。我们的代码和微调模型可在https://github.com/YoungSeng/Egocentric-Co-Pilot获取。

英文摘要:

Multimodal Large Language Models (MLLMs) struggle with complex video QA benchmarks like HD-EPIC VQA due to ambiguous queries/options, poor long-range temporal reasoning, and non-standardized outputs. We propose a framework integrating query/choice pre-processing, domain-specific Qwen2.5-VL fine-tuning, a novel Temporal Chain-of-Thought (T-CoT) prompting for multi-step reasoning, and robust post-processing. This system achieves 41.6% accuracy on HD-EPIC VQA, highlighting the need for holistic pipeline optimization in demanding video understanding. Our code, fine-tuned models are available at https://github.com/YoungSeng/Egocentric-Co-Pilot.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑