ArlingtonCL2/Barkopedia_Dog_Act_Env

Barkopedia挑战数据集提供了一个标记好的狗吠音频剪辑数据集，用于从声音中理解活动和环境。数据集包括训练集和即将发布的测试集，每个剪辑都标注有一个活动类别和一个环境类别。活动类别包括休息、对声音警觉、寻求注意、与人类玩耍、与其他动物玩耍、与玩具玩耍、乞讨食物、洗澡等。环境类别包括室内（一般）、靠近窗户、靠近门、在草地上、靠近其他动物、车内等。数据集的标签最初是通过视频辅助推理和视觉语言模型（Janus-Pro-7B）生成的，并经过人工验证以确保质量。

The Barkopedia Challenge Dataset provides a labeled dataset of dog bark audio clips for understanding activity and environment from sound. It includes a training set with 12,480 audio clips and a test set to be released in June with 3,120 audio clips. Each clip is annotated with one activity category and one environment category. Activity categories include resting, alerting to sounds, seeking attention, playing with humans, playing with other animals, playing with toys, begging for food, and taking a shower. Environment categories include indoor (general), near a window, near a door, on grass, near other animals, and vehicle interior. The labels were initially generated using video-assisted inference via a visual-language model (Janus-Pro-7B) and later manually verified for quality assurance.