Stellar

该数据集名为“Stellar”，为基于人类个性化文本生成图像的任务提供了一个大规模的标准化评估数据集。它包含了2万个精心挑选的提示，描述了富有想象力的情境和动作，以渲染以人为中心的虚构图像。此外，该数据集还包含了丰富的元数据注释，用于评估目的，提示语由专家标注员精心挑选，并通过半自动生成过程来增强多样性。数据集的规模为2万个提示和400个独特的人类身份。其任务目标是个性化文本到图像的生成。

The dataset named "Stellar" is a large-scale standardized evaluation dataset for personalized human-centric text-to-image generation tasks. It contains 20,000 carefully curated prompts that depict imaginative scenarios and actions for rendering human-focused fictional images. Additionally, the dataset includes rich metadata annotations for evaluation purposes. The prompts are meticulously selected by expert annotators, and their diversity is enhanced via a semi-automated generation process. The dataset comprises 20,000 prompts and 400 unique human identities, with its targeted task being personalized text-to-image generation.