creative-graphic-design/BannerRequest400

BannerRequest400是第一个专门设计用于评估广告横幅生成系统的多模态基准数据集。该数据集填补了现有设计数据集的空白，提供了视觉和文本输入。数据集包含100个独特的品牌标志（PNG和SVG格式）和400个不同的横幅设计请求（文本模态）。标志使用Claude 3.5 Sonnet合成生成，并通过专家细化以确保真实美学和避免偏见。研究人员利用GPT-4o将每个设计意图扩展为四个不同的横幅请求，针对不同的受众-目的组合，从而产生400个多样化的规格，并进一步扩展到13个标准横幅尺寸，形成一个包含5,200个多模态横幅规格的综合评估集。该基准首次实现了对横幅生成方法在不同设计请求、多个上下文（不同受众细分和活动目的）以及行业标准显示尺寸上的严格评估。

BannerRequest400 is the first multimodal benchmark specifically designed to evaluate advertising banner generation systems. It addresses gaps in existing design datasets by providing both visual and textual inputs. The dataset pairs 100 unique brand logos (provided in both PNG and SVG formats) with 400 diverse banner design requests (textual modality). The logos were synthetically generated using Claude 3.5 Sonnet, then refined by experts to ensure authentic aesthetics and avoid bias. Researchers leveraged GPT-4o to expand each design intention into four distinct banner requests targeting different audience-purpose combinations. This systematic expansion yielded 400 diverse specifications, further extended across 13 standard banner dimensions for a comprehensive evaluation set of 5,200 multimodal banner specifications. The benchmark enables the first rigorous evaluation of banner generation approaches across diverse design requests, multiple contexts (different audience segments and campaign purposes), and industry-standard display sizes.