CookGen
收藏资源简介:
CookGen是一个专注于烹饪领域的大规模视频数据集,由约翰斯·霍普金斯大学和字节跳动Seed团队创建。该数据集包含约200,000个视频片段,每个片段平均时长为9.5秒,数据来源于YouCook2和HowTo100M。数据集经过严格的质量过滤和字幕匹配处理,确保每个视频片段都具有清晰的叙事流程和详细的字幕描述。CookGen旨在支持长叙事视频生成的研究,特别是在烹饪领域,通过提供结构化的视频数据,帮助模型生成具有视觉和语义一致性的长视频。该数据集的应用领域包括视频生成、自然语言处理以及多模态学习等。
CookGen is a large-scale video dataset dedicated to the culinary domain, developed jointly by Johns Hopkins University and the ByteDance Seed Team. Comprising approximately 200,000 video clips with an average duration of 9.5 seconds each, the dataset is sourced from YouCook2 and HowTo100M. Rigorous quality filtering and subtitle matching procedures have been applied to ensure that every video clip features a clear narrative flow and detailed subtitle descriptions. CookGen aims to support research on long-form narrative video generation, particularly in the culinary domain, by providing structured video data to assist models in generating long videos with visual and semantic consistency. Potential application scenarios of this dataset include video generation, natural language processing, and multimodal learning, among others.

- 1VideoAuteur: Towards Long Narrative Video Generation约翰斯·霍普金斯大学, 字节跳动Seed · 2025年



