T2I-FactualBench
收藏资源简介:
T2I-FactualBench是由浙江大学和阿里巴巴集团共同创建的一个大规模数据集,旨在评估文本到图像生成模型在处理知识密集型概念时的准确性。该数据集包含1600个知识密集型概念,涵盖8个不同领域,如动物、人工制品、食物等。数据集的创建过程包括从知识库中筛选和收集知识密集型概念,并通过三层任务框架评估模型的生成能力。T2I-FactualBench主要应用于评估和提升文本到图像生成模型在复杂知识概念生成中的准确性和可靠性。
T2I-FactualBench is a large-scale dataset co-created by Zhejiang University and Alibaba Group, which aims to evaluate the accuracy of text-to-image generation models when handling knowledge-intensive concepts. This dataset contains 1,600 knowledge-intensive concepts spanning 8 distinct domains such as animals, artifacts, food, etc. The creation process of the dataset includes screening and collecting knowledge-intensive concepts from knowledge bases, and evaluating the generation capabilities of models through a three-tier task framework. T2I-FactualBench is primarily applied to evaluate and enhance the accuracy and reliability of text-to-image generation models in generating complex knowledge-based concepts.




