Kinetics 700

Kinetics-700 是一个包含 650,000 个剪辑的视频数据集，涵盖 700 个人类动作类别。 这些视频包括弹奏乐器等人与物之间的互动，以及握手和拥抱等人与人之间的互动。 每个动作类至少有 700 个视频片段。 每个剪辑都标有一个动作类，持续大约 10 秒。

Kinetics-700 is a video dataset consisting of 650,000 video clips covering 700 human action categories. These videos include interactions between humans and objects such as playing musical instruments, as well as interpersonal interactions like handshakes and hugs. Each action category contains no fewer than 700 video clips. Each clip is labeled with one action category and has a duration of approximately 10 seconds.

搜集汇总

背景与挑战

背景概述

Kinetics 700是一个包含650,000个视频剪辑的大规模数据集，涵盖700个人类动作类别，每个动作类至少有700个视频片段，每个剪辑持续约10秒并标有动作类别。该数据集由DeepMind发布，适用于行为识别等计算机视觉任务。