Manga109Story
收藏资源简介:
Manga109Story数据集由美团研究团队构建,旨在支持基于纯文本的漫画生成研究。该数据集基于Manga109数据集,包含21142页由专业日本漫画家绘制的漫画,涵盖了从1970年代到2010年代的多种读者群体和题材。数据集通过多模态大语言模型(MLLM)为每页漫画生成面板描述和故事摘要,提供了丰富的标注信息,如面板坐标、角色名称、对话内容等。该数据集的应用领域主要集中在漫画生成任务,旨在通过纯文本生成多面板漫画页面,解决漫画创作中的连贯性、布局合理性和角色一致性等问题。
The Manga109Story dataset was constructed by the Meituan Research Team, aiming to support pure-text-based comic generation research. Built upon the Manga109 dataset, it contains 21,142 comic pages drawn by professional Japanese cartoonists, covering diverse readership groups and genres spanning from the 1970s to the 2010s. Leveraging multimodal large language models (MLLMs), the dataset generates panel descriptions and story summaries for each comic page, and provides rich annotation information such as panel coordinates, character names, dialogue content, and more. The main application field of this dataset focuses on comic generation tasks, which aims to generate multi-panel comic pages through pure text and address core challenges in comic creation including coherence, reasonable layout, and character consistency.




