7Bench

7Bench是一个全面评估布局引导文本到图像模型性能的基准测试数据集，由意大利帕多瓦大学数学系和意大利都灵Brain Technologies srl公司大脑技术部门共同创建。该数据集包含224个样本，每个样本由文本提示和一组边界框组成，用于评估模型在不同场景下的生成能力，包括对象生成、颜色保真度、属性识别、对象间关系和空间控制。7Bench旨在解决现有基准测试数据集在评估布局引导模型时对布局对齐性评估不足的问题，通过引入布局对齐分数来评估空间精度。该数据集为评估和比较不同布局引导扩散模型提供了统一的标准，有助于推动相关领域的研究和发展。

7Bench is a benchmark dataset for comprehensively evaluating the performance of layout-guided text-to-image models, co-developed by the Department of Mathematics of the University of Padua in Italy and the Brain Technologies division of Brain Technologies srl based in Turin, Italy. This dataset comprises 224 samples, each consisting of a text prompt and a set of bounding boxes, designed to evaluate the generative capabilities of models across various scenarios, including object generation, color fidelity, attribute recognition, inter-object relationships and spatial control. 7Bench targets the shortcoming of insufficient layout alignment evaluation in existing benchmark datasets when assessing layout-guided text-to-image models, and introduces layout alignment scores to quantify spatial accuracy. This dataset provides a unified standard for evaluating and comparing different layout-guided diffusion models, which contributes to promoting research and development in related fields.