data-is-better-together/open-image-preferences-v1-results

该数据集旨在创建10K个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。数据集通过社区注释生成，包含图像对和相应的提示。数据集可以通过Argilla或HuggingFace的`datasets`库加载。

The dataset aims to create 10,000 text-to-image preference pairs to evaluate the performance of image generation models across a variety of common image categories, based on prompts with varying levels of difficulty. The creation process involves using prompts from [fal/imgsys-results], evolving them based on complexity and quality, and then annotating preferences between generated images through community feedback. The result is 10,000 annotated preference pairs. The dataset can be loaded and used with Argilla or the `datasets` library. The dataset structure includes fields, questions, metadata, vectors, and data instances. Detailed guidelines for dataset creation and annotation are provided, including definitions for the image preference task and annotation shortcuts.