DesignSense：用于图形布局生成的人类偏好数据集和奖励建模框架

Graphic layouts play a crucial role as a medium for visual communication across various channels. Recent layout generation models have shown impressive capabilities; however, they often fail to align with nuanced human aesthetic judgments. Existing preference datasets and reward models trained for text-to-image generation do not generalize well to layout evaluation, where the spatial arrangement of identical elements determines quality. To address this critical gap, DesignSense-10k is introduced, comprising a large-scale dataset of 10,235 human-annotated preference pairs for graphic layout evaluation. A five-stage curation pipeline is proposed to generate visually coherent layout transformations across diverse aspect ratios, utilizing semantic information for optimization.