MangaZero 漫画图像数据集

MangaZero 数据集是一个专为漫画生成任务设计的大规模、多角色、多状态的漫画图像数据集，由北京大学、上海人工智能实验室、南洋理工大学和字节跳动于 2024 年发布，相关论文成果为「DiffSensei: Bridging Multi-Modal LLMs and Diffusion Models for Customized Manga Generation」。具体包含 43,264 页漫画和 427,147 个带注释的面板，支持在连续帧中可视化各种角色互动和动作，适用于多角色、多状态的漫画生成任务。

The MangaZero dataset is a large-scale, multi-character and multi-state comic image dataset specifically designed for comic generation tasks. It was released in 2024 by Peking University, Shanghai AI Laboratory, Nanyang Technological University and ByteDance, with its associated research paper titled "DiffSensei: Bridging Multi-Modal LLMs and Diffusion Models for Customized Manga Generation". The dataset specifically contains 43,264 comic pages and 427,147 annotated panels, which enables the visualization of various character interactions and actions across consecutive frames, and is suitable for multi-character and multi-state comic generation tasks.

搜集汇总

背景与挑战

背景概述

MangaZero 漫画图像数据集由北京大学、上海人工智能实验室、南洋理工大学和字节跳动于2024年发布，相关论文为'DiffSensei: Bridging Multi-Modal LLMs and Diffusion Models for Customized Manga Generation'。该数据集包含43,264页漫画和427,147个带注释的面板，适用于多角色、多状态的漫画生成任务，支持图像描述和图像分类。