Rapidata/117k_human_preferences_flux1.0_V_flux1.1Blueberry

Rapidata图像生成对齐数据集是一个包含人类对AI生成图像比较评价的数据集。它由340k人类注释数据集的三分之一组成，分为偏好、连贯性和文本到图像对齐三种模态。数据集重点关注由Flux 1.0和Flux 1.1Blueberry生成的图像对，参与者被要求选择他们更喜欢的图像。该数据集拥有超过117,000个单独的人类偏好投票，涵盖了281个精心设计的提示，用于测试图像生成的各个方面。

The Rapidata Image Generation Alignment Dataset is a dataset of human comparative evaluations of AI-generated images. It consists of one-third of a 340k human annotation dataset, split into three modalities: Preference, Coherence, and Text-to-Image Alignment. The dataset focuses on image pairs generated by Flux 1.0 and Flux 1.1Blueberry, with participants asked to select their preferred image. The dataset includes over 117,000 individual human preference votes across 281 carefully curated prompts designed to test various aspects of image generation.