HowToDIV
收藏资源简介:
HowToDIV数据集是一个大规模的多轮对话数据集,包含507个对话,6636个问答对和24小时的视频剪辑,涵盖了烹饪、机械和种植等多样化的任务。数据集由一个专家和一个新手用户之间的真实对话组成,其中专家一步一步地教授用户如何完成任务。每个对话都伴随着用户视角的视频剪辑,以及每段对话的指令步骤。数据集旨在解决现实世界中任务辅助的需求,为AI代理提供语言视觉模型和代理,以帮助用户完成复杂的、多步骤的任务。
The HowToDIV dataset is a large-scale multi-turn dialogue dataset, containing 507 dialogues, 6636 question-answer pairs, and 24 hours of video clips, covering diverse tasks such as cooking, mechanical operations, and planting. The dataset consists of real dialogues between an expert and a novice user, where the expert teaches the user how to complete the task step by step. Each dialogue is accompanied by video clips from the user's perspective, as well as the instructional steps for each dialogue segment. This dataset aims to address the demand for real-world task assistance, providing language-vision models and AI Agents with resources to help users complete complex, multi-step tasks.




