Key-step Skill Generation Dataset
收藏资源简介:
本数据集由南京大学State Key Laboratory for Novel Software Technology构建,包含约110,000个子剪辑,旨在支持关键步骤技能生成任务(KS-Gen)。数据集通过综合处理COIN、CrossTask和HT-Step三个原始数据集,并对视频片段进行剪辑、优化和描述,以适应技能生成中的关键步骤。数据集涵盖了多个领域,如家务、车辆、护理等,包含了多个不同技能的操作视频,为长视频生成任务提供了高质量的训练数据。
This dataset was constructed by the State Key Laboratory for Novel Software Technology at Nanjing University, comprising approximately 110,000 sub-clips, and is specifically designed to support the Key Step Skill Generation (KS-Gen) task. It is developed by comprehensively processing three original datasets: COIN, CrossTask, and HT-Step, followed by clipping, optimizing, and annotating the video segments to align with the requirements of key step-oriented skill generation. The dataset covers multiple domains including household chores, vehicle-related operations, caregiving, and more, and contains instructional videos of various distinct skills, providing high-quality training data for long-form video generation tasks.




