PICD: Photographic Image Composition Dataset
收藏资源简介:
PICD是一个用于摄影图像构图分析的大规模数据集,目前包含49,123张高质量图像,标注了24种构图类别。该数据集旨在支持AI模型中构图学习的评估和进展,适用于美学质量评估、构图感知图像裁剪等多种任务。
PICD is a large-scale dataset for photographic image composition analysis. Currently, it comprises 49,123 high-quality images annotated with 24 composition categories. This dataset aims to support the evaluation and advancement of composition learning in AI models, and is applicable to various tasks such as aesthetic quality assessment and composition-aware image cropping.
PICD: Photographic Image Composition Dataset 概述
📌 数据集简介
- 名称: PICD (Photographic Image Composition Dataset)
- 目的: 支持AI模型中的构图学习评估与推进
- 应用场景: 美学质量评估、构图感知图像裁剪等
- 论文: CVPR 2025 Paper
- 补充材料: Supplementary Appendix
📊 数据集内容
- 图像数量: 49,123张高质量图像
- 标注信息: 24种构图类别
- 标签系统:
- 元素类型: 点、线、形状(基于康定斯基原理)
- 排列模式: 三分法则、居中、对角线、垂直、水平、三角形、C曲线、O曲线、S曲线、放射状、密集、分散等
🔗 下载信息
1. 图像
- 可直接下载部分: 44,577张(来自Unsplash、Pexels等开放平台)
- 需自行获取部分: 4,546张(来自AVA等不允许重新分发的公共数据集)
- 替代下载方式: 通过邮件申请(picd2025@outlook.com)
2. 标注
- 标注文件下载
- 包含字段:
img_id,category_id,category_abbre,category_full_name
- 包含字段:
📄 许可与条款
- 许可证: CC BY-NC-SA 4.0
- 使用条款: PICD Dataset Terms of Use
🔧 引用
bibtex @inproceedings{zhao2025can, title={Can Machines Understand Composition? Dataset and Benchmark for Photographic Image Composition Embedding and Understanding}, author={Zhao, Zhaoran and Lu, Peng and Zhang, Anran and Li, Peipei and Li, Xia and Liu, Xuannan and Hu, Yang and Chen, Shiyi and Wang, Liwei and Guo, Wenhao}, booktitle={Proceedings of the Computer Vision and Pattern Recognition Conference}, pages={14411--14421}, year={2025} }




