CelebV-HQ: A Large-Scale Video Facial Attributes Dataset

Large-scale datasets have played indispensable roles in the recent success of face generation/editing and significantly facilitated advances emerging research fields. However, academic community still lacks a video dataset with diverse facial attribute annotations, which is crucial for on face-related videos. In this work, we propose large-scale, high-quality, rich named High-Quality Celebrity Video Dataset (CelebV-HQ). CelebV-HQ contains 35, 666 clips resolution $$512\times 512$$ at least, involving 15, 653 identities. All are labeled manually 83 attributes, covering appearance, action, emotion. We conduct comprehensive analysis terms age, ethnicity, brightness stability, motion smoothness, head pose diversity, data quality to demonstrate diversity temporal coherence CelebV-HQ. Besides, its versatility potential validated two representative tasks, i.e., unconditional generation editing. finally envision future CelebV-HQ, as well new opportunities challenges it would bring related directions. Data, code, models publicly available (Project page: https://celebv-hq.github.io/ Code models: https://github.com/CelebV-HQ/CelebV-HQ ).