creative-graphic-design/GraphicDesignEvaluation

GraphicDesignEvaluation数据集用于评估大型多模态模型（如GPT-4o）是否能够根据核心设计原则（特别是对齐、重叠和空白）评估图形设计的质量。该数据集包含700个横幅和海报设计（100个原始和600个扰动），每个设计由60个人工标注者进行评分。每个图像都有相应的人类评分（1-10分）和基于GPT的三个原则的评分，从而可以研究人类判断、启发式指标和基于LMM的评估之间的相关性。数据集以英文编写，并遵循Apache-2.0许可证。

The GraphicDesignEvaluation dataset is designed to evaluate the ability of large multimodal models (LMMs) to assess the quality of graphic designs based on core design principles—specifically alignment, overlap, and white space. It contains 700 banner and poster designs (100 original and 600 perturbed), each rated by 60 human annotators. Each image has associated human scores (1–10 scale) and GPT-based scores for the three principles, enabling the study of correlations between human judgment, heuristic metrics, and LMM-based evaluation. The dataset is in English and is licensed under Apache-2.0.