Rapidata/open-image-preferences-v1-more-results-binarized

该数据集旨在创建10K文本到图像的偏好对，用于评估图像生成模型在不同图像类别中的表现。通过使用来自fal/imgsys-results的提示，并根据复杂性和质量进行演变，社区成员被要求对每个提示生成的两张图像进行偏好标注。Rapidata收集了来自全球的超过170,000个个人偏好，每个图像对大约有10个偏好标注。标注者被要求根据描述选择他们偏好的图像，每个图像对被展示给10个标注者，图像的位置是随机打乱的。数据集的格式是二值化的，对于每个图像对，计算了用户评分加权的评分，并列出选择的模型和拒绝的模型。

The dataset aims to create 10,000 text-to-image preference pairs for evaluating the performance of image generation models across a wide variety of common image categories. Using prompts from [fal/imgsys-results](https://huggingface.co/datasets/fal/imgsys-results), the community was asked to annotate preferences between two generated images for each prompt. Rapidata collected over 170,000 individual preferences from people around the globe. The dataset is binarized, with each pair having a user-score weighted ratio, listing a chosen and rejected model.