CelebV-HQ

CelebV-HQ包含35,666个视频剪辑，涉及15,653个身份和83个手动标记的面部属性，涵盖外观，动作和情感。我们对种族，年龄，亮度，运动平滑度，头部姿势多样性和数据质量进行了全面分析，以证明celebv-hq的多样性和时间一致性。此外，它的通用性和潜力在无条件视频生成和视频面部属性编辑任务上得到验证。

CelebV-HQ contains 35,666 video clips, involving 15,653 unique identities and 83 manually annotated facial attributes covering appearance, actions and emotions. We performed a comprehensive analysis across ethnicity, age, brightness, motion smoothness, head pose diversity and data quality to validate the diversity and temporal consistency of CelebV-HQ. Furthermore, its generality and potential have been verified on the tasks of unconditional video generation and video facial attribute editing.

搜集汇总

背景与挑战

背景概述

CelebV-HQ是一个大规模人脸视频数据集，包含35,666个视频剪辑，覆盖15,653个不同身份和83个手动标记的面部属性，涵盖外观、动作和情感等多维度特征。该数据集经过全面分析，确保在种族、年龄、运动平滑度等方面的多样性和时间一致性，适用于无条件视频生成和视频面部属性编辑等计算机视觉任务。