FineCapEval

该数据集名为FineCapEval，是一个用于评估字幕的全新数据集，其特点在于采用了细粒度的评估标准，包括整体、背景、物体和关系等四个方面。该数据集由1,000张图片组成，总共包含5,000个跨越这四个标准的注释。在构建此数据集时，研究人员收集了图片，并让人类标注员编写包含背景、物体、关系以及一个综合描述的详细字幕。规模上，数据集包含了1,000张图片和5,000个注释，其任务是进行图像字幕评估。

This dataset, named FineCapEval, is a novel dataset for image captioning evaluation, which features fine-grained evaluation criteria covering four dimensions: overall quality, background, objects, and relational details. The dataset consists of 1,000 images and a total of 5,000 annotations spanning these four criteria. During its construction, researchers collected the images and had human annotators write detailed captions that include background information, object descriptions, relational details, and a comprehensive overall description. In terms of scale, the dataset contains 1,000 images and 5,000 annotations, and its core task is image captioning evaluation.