DH-FaceEmoVid-150
收藏资源简介:
DH-FaceEmoVid-150数据集是由浙江实验室、理想汽车、哈尔滨工业大学、浙江大学和中国科学院大学杭州高等研究院联合创建的高分辨率(1080p)视频数据集,专门用于人类面部情感表达的研究。该数据集包含150小时的视频内容,涵盖了六种基本情绪(愤怒、厌恶、恐惧、快乐、悲伤、惊讶)和四种复合情绪(愤怒厌恶、悲伤惊讶、悲伤恐惧、快乐惊讶)。数据集的创建过程包括从多种来源收集数据,并通过情感强度过滤、动作单元(AU)标签提取和文本指令生成等步骤进行优化。该数据集的应用领域主要集中在音频驱动的肖像动画生成,旨在解决现有模型在生成复杂情感和细微面部表情方面的局限性,推动情感控制技术的发展。
DH-FaceEmoVid-150 is a high-resolution (1080p) video dataset jointly created by Zhejiang Lab, Li Auto, Harbin Institute of Technology, Zhejiang University, and Hangzhou Institute for Advanced Study, University of Chinese Academy of Sciences, specifically for research on human facial emotional expressions. This dataset contains 150 hours of video content, covering six basic emotions (anger, disgust, fear, happiness, sadness, surprise) and four composite emotions (anger-disgust, sadness-surprise, sadness-fear, happiness-surprise). The creation process of the dataset includes collecting data from diverse sources, and optimizing it through steps such as emotional intensity filtering, action unit (AU) label extraction and text instruction generation. Its application fields mainly focus on audio-driven portrait animation generation, aiming to address the limitations of existing models in generating complex emotions and subtle facial expressions, and advance the development of emotion control technologies.

- 1MoEE: Mixture of Emotion Experts for Audio-Driven Portrait Animation浙江实验室, 理想汽车, 哈尔滨工业大学, 浙江大学, 中国科学院大学杭州高等研究院 · 2025年



