Automatic classification of cichlid behaviors using 3D Convolutional Residual Networks. Lijiang Long

This dataset contains 14172 short video clips which recorded cichlids behaviors including feeding and bower building. Along with the clips, there is an annotation file 'AnnotationFile.csv' where each video clip is annotated by an experienced human observer. In short, each video clip was classified into one of the ten categories: Bower scoop (c), Bower spit (p), Bower multiple (b), Feeding scoop (f), Feeding spit (t), Feeding multiple (m), Spawn/quiver (s), Sand dropping (d), Other (o), Shadow/reflection (x).

本数据集包含14172段短视频片段，记录了慈鲷（cichlids）的取食与筑巢行为。随数据集附带标注文件"AnnotationFile.csv"，所有视频片段均由资深人类观察员完成标注。简言之，每段视频片段被划分为以下十大类别：筑巢铲挖（Bower scoop，c）、筑巢吐沙（Bower spit，p）、多步筑巢行为（Bower multiple，b）、取食铲挖（Feeding scoop，f）、取食吐沙（Feeding spit，t）、多步取食行为（Feeding multiple，m）、产卵/震颤（Spawn/quiver，s）、沙粒抛撒（Sand dropping，d）、其他（Other，o）、阴影/倒影（Shadow/reflection，x）。