SorrowTea/PhotoBench

PhotoBench是一个基于真实个人相册构建的图像检索基准数据集，旨在从视觉匹配转向个性化的多源意图驱动的照片检索。该数据集包含3个相册的测试查询（中英文）和原始图像，支持开放式的自然语言查询，允许用户使用自己的嵌入模型、标题生成器或基于代理的检索工作流程。原始图像由于大小限制未包含在仓库中，需单独下载。

PhotoBench is an image retrieval benchmark constructed from authentic, personal albums, designed to shift the paradigm from visual matching to personalized multi-source intent-driven photo retrieval. The dataset includes test queries (in both English and Chinese) and raw images for 3 albums, supporting open-ended natural language queries. It allows users to employ their own embedding models, caption generators, or agent-based retrieval workflows. Raw images are not included in the repository due to size constraints and must be downloaded separately.