Affectiva's large-scale dataset

本研究使用的是由Affectiva公司提供的大规模数据集，包含约55,000个视频，这些视频记录了来自全球90多个国家的参与者在自然环境下的面部表情。数据集内容丰富，涵盖了多种族群和性别，视频中的动作单元（AU）由经过训练的FACS编码员手动标注。数据集的创建旨在支持自动面部表情识别（FER）和动作单元（AU）检测的研究，特别是在广告测试、驾驶员状态监控和社交机器人等领域。通过此数据集，研究者可以评估不同的预处理、分类和训练设置对AU检测性能的影响，从而优化模型设计。

The large-scale dataset used in this study is provided by Affectiva, which contains approximately 55,000 videos recording naturalistic facial expressions from participants across more than 90 countries worldwide. This dataset is comprehensive, covering diverse ethnic groups and genders. The Action Units (AUs) in the videos were manually annotated by trained FACS-certified coders. This dataset was developed to support research on automatic facial expression recognition (FER) and action unit (AU) detection, particularly in applications such as advertising testing, driver state monitoring, and social robotics. With this dataset, researchers can evaluate the impact of various preprocessing, classification, and training configurations on AU detection performance, thereby optimizing model design.

搜集汇总

背景与挑战

背景概述

该数据集是一个大规模的面部动作单元检测数据集，包含约55,000个在自然环境中收集的视频，用于研究不同预处理和分类训练设置对AU检测的影响。