Short Film Dataset (SFD)

Short Film Dataset (SFD) 是由国立高等先进工业科学研究所创建的一个包含1,078部公开可访问的业余电影的数据集，总时长超过243小时，平均每部电影时长13分钟。SFD涵盖多种电影类型，如剧情、科幻、恐怖等，每部电影都附有简短的剧情概要（logline）和详细描述，确保数据集内容丰富且无数据泄露问题。数据集创建过程中，利用GPT-4自动生成问题-答案对，并经过人工审核确保质量。SFD主要用于长篇故事级视频理解的研究，旨在解决现有数据集在视频时长、内容丰富度和数据泄露方面的局限性。

The Short Film Dataset (SFD) is a collection of 1,078 publicly accessible amateur films developed by the National Institute of Advanced Industrial Science and Technology. It has a total runtime of over 243 hours, with an average duration of 13 minutes per film. SFD covers a diverse range of film genres, including drama, science fiction, horror, and others. Each film in the dataset is paired with a brief logline and detailed synopsis, ensuring the dataset is rich in content and free of data leakage issues. During its development, GPT-4 was used to automatically generate question-answer pairs, which were manually reviewed to ensure quality. SFD is primarily intended for research on long-form story-level video understanding, aiming to address the limitations of existing datasets in terms of video duration, content richness, and data leakage.