Rapidata/human-coherence-preferences-images

Rapidata图像生成连贯性数据集是用于文本到图像模型连贯性评估的最大人类注释数据集之一，包含了超过1,200,000个人类连贯性投票。该数据集通过Rapidata Python API在短短4天内收集，覆盖了全球145多个国家的参与者。数据集的特点包括大规模、全球代表性、多样化的提示和领先的模型比较。数据集的应用场景包括基准测试新模型、开发更好的评估指标、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。

The Rapidata Image Generation Coherence Dataset is one of the largest human-annotated coherence datasets for text-to-image models, containing over 1,200,000 human coherence votes. The dataset was collected in just 4 days using the Rapidata Python API, with participants from over 145 countries worldwide. Key features of the dataset include massive scale, global representation, diverse prompts, and comparisons between leading models. Applications of the dataset include benchmarking new models, developing better evaluation metrics, understanding global preferences in AI-generated imagery, training and fine-tuning image generation models, and researching cross-cultural aesthetic preferences.