BabyRobot Emotion Database (BRED)

BabyRobot Emotion Database (BRED) This dataset includes extracted data from videos of children performing emotions. The dataset has a total of 215 samples and includes: Skeletons extracted by OpenPose. Facial landmarks extracted by the [OpenFace toolkit](OpenFace toolkit) Features extracted by a ResNet 50 network pretrained in the AffectNet database. Features are in PyTorch format. Annotations by three different annotators. The annotations are hierarchical and apart from the ground truth emotion, denote if the child used its body and/or face for performing the emotion. <strong>Note</strong> that there is an error in the annotations.csv file. Replace "spontaneous" in the path with "game" and "acted" with "pre-game" to get the correct paths. We also provide the code on Github. The accompanied paper can be found on arXiv.BabyRobot情感数据库（BabyRobot Emotion Database, BRED） 本数据集提取自儿童情绪表演视频，总计包含215个样本，具体涵盖以下内容： 1. 由OpenPose提取的人体骨骼关键点数据； 2. 由OpenFace工具包提取的面部关键点； 3. 由在AffectNet数据库上预训练的ResNet 50网络提取的特征，该类特征采用PyTorch格式存储； 4. 由三名不同标注人员完成的标注信息。 该标注采用层级化结构，除标注情绪真值外，还会标记该儿童是否通过肢体或面部表情来完成情绪表演。 **注意**：annotations.csv文件中存在一处错误，请将路径中的"spontaneous"替换为"game"，将"acted"替换为"pre-game"，即可获取正确的文件路径。 我们还在Github平台上提供了配套代码，相关配套论文可在arXiv平台上查阅。