EVENT DOSSIER
On September 7, 2026, the EyeMakeYou model was released on the arXiv cs.CV platform. This research proposes a framework based on diffusion models aimed at synthesizing eye movement trajectories with high frequency. Unlike existing methods, EyeMakeYou can generate data based on three conditions: identity, task type, and subjective state. This allows it to effectively capture the impact of individual differences and situational changes on gaze patterns while maintaining high-frequency eye movement data.
EyeMakeYou 提出了一种多条件去噪扩散框架，用于生成身份去除后的参考轨迹对应的 5 秒、1000Hz 双变量眼动速度序列。该方法在去噪过程中结合身份嵌入、任务嵌入以及用户自报告的整体难度、精神疲劳和眼部疲劳评分进行条件控制，并融合扩散噪声预测与身份保持目标，同时引入多分辨率频谱、漂移一致性及事件加权局部平滑损失函数。基于 GazeBase 数据集的实验表明，EyeMakeYou 在空间精度和特征空间的真实 - 合成相似度上优于现有生成方法，且保留了主观报告与眼动特征间的任务依赖关联，支持条件扩散技术用于增强生物识别及交互应用的眼动数据集。