Rapidata/human-alignment-preferences-images

Rapidata Image Generation Alignment数据集是一个大规模的人类注释对齐数据集，包含超过1,200,000个人类偏好投票，用于比较不同文本到图像生成模型的表现。数据集的特征包括提示、两张图像、每张图像的投票数、生成图像的模型、详细结果以及图像路径。数据集的应用包括基准测试新图像生成模型、开发更好的生成模型评估指标、理解全球对AI生成图像的偏好、训练和微调图像生成模型以及研究跨文化审美偏好。数据集通过Rapidata的创新注释平台在不到100小时内收集完成，覆盖了145多个国家，具有全球代表性和大规模数据收集能力。

The Rapidata Image Generation Alignment Dataset is a large-scale human-annotated alignment dataset containing over 1,200,000 human preference votes for comparing the performance of different text-to-image generation models. The dataset features include prompts, two images, votes for each image, the models that generated the images, detailed results, and image paths. Applications of the dataset include benchmarking new image generation models, developing better evaluation metrics for generative models, understanding global preferences in AI-generated imagery, training and fine-tuning image generation models, and researching cross-cultural aesthetic preferences. The dataset was collected in under 100 hours through Rapidatas innovative annotation platform, covering over 145 countries, with global representation and large-scale data collection capabilities.