P4I (PoemForImage)
收藏资源简介:
P4I数据集是由印度理工学院帕特纳分校计算机科学与工程学院和Adobe Research合作创建的,包含1111首诗歌,涵盖多种主题,如叙事诗、自由诗、颂歌、哀歌、开放诗、十四行诗、开放式对句和儿童诗歌。数据集旨在促进诗歌向图像生成的转化研究,通过引入多阶段提示精炼循环,提高诗歌文本到图像转换过程中的信息保留,生成能够有效捕捉诗歌本质和意义的图像。数据集的创建涉及了从在线和离线资源中收集诗歌,并进行了结构化元数据标注,以改善诗歌组织和促进图像生成。数据集的引入为诗歌图像生成领域提供了新的研究方向,并鼓励了未来对诗歌研究的探索。
The P4I Dataset was co-created by the Department of Computer Science and Engineering, Indian Institute of Technology Patna, and Adobe Research. It comprises 1,111 poems covering diverse genres including narrative poetry, free verse, odes, elegies, open-form poetry, sonnets, open-form couplets, and children's poetry. This dataset aims to advance research on poem-to-image generation, by introducing a multi-stage prompt refinement loop to enhance information retention during the text-to-image conversion process, thereby generating images that effectively capture the essence and intended meaning of the original poems. The creation of the dataset involved collecting poems from both online and offline resources, and conducting structured metadata annotation to improve poem organization and facilitate image generation. The release of this dataset provides new research directions for the field of poem-to-image generation, and encourages future explorations in poetry-related research.
PoemTale-Diffusion 数据集概述
数据集简介
- 数据集名称:P4I (PoemForImage)
- 数据量:1111首诗歌
- 数据来源:多个线上和线下资源
- 主要用途:支持诗歌到图像生成的研究
数据集特点
- 诗歌类型:包含复杂、抽象或高度描述性语言的诗歌
- 数据质量:经过诗歌专家小组的定性评估
- 应用场景:用于测试和改进诗歌到图像生成模型的信息保留能力
相关研究背景
- 研究课题:诗歌到图像生成中的信息保留问题
- 技术挑战:现有模型难以处理具有多层次、抽象和双重含义的诗歌语言
- 解决方案:采用多阶段提示细化循环和一致自注意力技术
数据集贡献
- 为诗歌到图像生成领域提供了专门的数据资源
- 支持对创造性语言处理能力的评估
- 促进文本到图像生成技术在艺术领域的应用研究

- 1PoemTale Diffusion: Minimising Information Loss in Poem to Image Generation with Multi-Stage Prompt Refinement印度理工学院帕特纳分校计算机科学与工程学院, Adobe Research · 2025年



