StoryReasoning
收藏资源简介:
StoryReasoning数据集是为了解决视觉叙事系统在保持角色身份一致性和将动作与适当主体关联方面的挑战而创建的。该数据集包含4,178个故事,这些故事是从52,016张电影图片中提取的,具有结构化的场景分析和有根据的故事。每个故事都保持了角色和对象在帧之间的一致性,并通过结构化的表格表示显式地建模多帧关系。该数据集的特点是使用视觉相似性和人脸识别进行跨帧对象重新识别,使用思维链推理进行明确的叙事建模,以及一种将文本元素与多个帧中的视觉实体链接起来的接地方案。该数据集旨在为视觉叙事系统提供一个新的基准,并解决现有系统在保持角色和对象一致性以及生成连贯叙事结构方面的局限性。
The StoryReasoning dataset was created to address the challenges faced by visual storytelling systems in maintaining consistent character identities and associating actions with their appropriate agents. This dataset includes 4,178 stories extracted from 52,016 film stills, featuring structured scene analysis and well-grounded narratives. Each story preserves the consistency of characters and objects across frames, and explicitly models inter-frame relationships via structured tabular representations. This dataset is characterized by three key aspects: leveraging visual similarity and facial recognition for cross-frame object re-identification, employing Chain-of-Thought (CoT) reasoning for explicit narrative modeling, and a grounding scheme that links textual elements to visual entities across multiple frames. This dataset aims to provide a novel benchmark for visual storytelling systems, and address the limitations of existing systems in maintaining consistent character and object identities as well as generating coherent narrative structures.




