HHOI

一个新的 RGB-D 视频数据集，即 UCLA Human-Human-Object Interaction (HHOI) 数据集，其中包括 3 种类型的人机交互，即握手、高五、拉起和 2 种人机交互人与物的互动，即扔、接、递杯子。平均而言，每个交互有 23.6 个实例，由来自不同视图的总共 8 个演员执行。每次交互持续 2-7 秒，以 10-15 fps 的速度呈现。A novel RGB-D video dataset, termed the UCLA Human-Human-Object Interaction (HHOI) dataset, is introduced. It encompasses three categories of human-human interactions: handshake, high-five, and arm-lifting, alongside two categories of human-object interactions involving throwing, catching, and cup handover. On average, each interaction type has 23.6 instances, which are performed by a total of 8 distinct actors across diverse viewpoints. Each interaction clip lasts between 2 and 7 seconds, and is recorded at a frame rate of 10 to 15 fps.

搜集汇总

背景与挑战

背景概述

HHOI是一个RGB-D视频数据集，专注于人-人交互（如握手、高五、拉起）和人-物交互（如扔、接、递杯子），平均每个交互包含23.6个实例，由8位演员从多视角录制，视频时长2-7秒，帧率为10-15 fps。该数据集由加州大学洛杉矶分校和印第安纳大学布卢明顿分校于2016年发布。