PosterDNA
收藏资源简介:
PosterDNA是由华南理工大学与南京信息工程大学联合开发的商业级海报生成数据集,包含16.6万条样本,首创HTML排版文件以支持可扩展文本渲染。该数据集包含三个专项子集:蓝图创建子集(5.7万条)通过逆向工程生成多粒度用户需求,图形生成子集(10万条)涵盖插画、极简等四种设计风格,统一文本布局子集(9000条)提供HTML结构化输出。数据来源于专业设计师协作构建的高质量海报样本,采用多阶段标注流程确保文本密度与布局复杂性。主要应用于商业海报自动生成领域,解决高密度小字号文本渲染不准和后期编辑困难等核心问题。
PosterDNA is a commercial-grade poster generation dataset jointly developed by South China University of Technology and Nanjing University of Information Science and Technology. It contains a total of 166,000 samples, and it is the first to adopt HTML layout files to support scalable text rendering. The dataset includes three specialized subsets: 1. The Blueprint Creation Subset (57,000 samples), which generates multi-granularity user requirements via reverse engineering; 2. The Graphic Generation Subset (100,000 samples) covering four design styles such as illustration and minimalism; 3. The Unified Text Layout Subset (9,000 samples) that provides HTML structured output. The data is derived from high-quality poster samples collaboratively built by professional designers, and a multi-stage annotation pipeline is employed to ensure control over text density and layout complexity. Its main applications lie in the field of automated commercial poster generation, addressing core issues such as inaccurate rendering of high-density small-font text and difficulties in post-editing.
PosterVerse数据集概述
数据集名称
PosterDNA
数据集简介
PosterDNA是首个面向商业级、文本密集的海报生成数据集,包含基于HTML的细粒度规范。该数据集旨在为模块化训练和验证提供高质量样本。
数据集特点
- 商业级质量:数据集样本符合商业级海报标准。
- 文本密集设计:海报设计包含密集的文本内容。
- 细粒度规范:提供基于HTML的详细规格说明。
- 用途:支持模块化训练和验证。
数据集状态
根据项目TODO列表,数据集尚未发布。
许可信息
代码和数据集应在非商业研究目的下,依据CC BY-NC-ND 4.0许可进行使用和分发。
版权声明
- 此存储库仅可用于非商业研究目的。
- 商业用途请联系Prof. Lianwen Jin (eelwjin@scut.edu.cn)。
- 版权归属:2026年,华南理工大学深度学习与视觉计算实验室。
相关论文
- 论文标题:PosterVerse: A Full-Workflow Framework for Commercial-Grade Poster Generation with HTML-Based Scalable Typography
- 作者:Junle Liu, Peirong Zhang, Yuyi Zhang, Pengyu Yan, Hui Zhou, Xinyue Zhou, Fengjun Guo, Lianwen Jin
- 会议:AAAI Conference on Artificial Intelligence
- 年份:2026
- 论文链接:https://arxiv.org/abs/2601.03993
联系方式
如有疑问,可通过junle_liu@foxmail.com联系Junle Liu。




