PKU-MMD

PKU-MMD是一种新的大规模基准，用于连续多模态3D人类动作理解，并涵盖了具有良好注释信息的广泛复杂人类活动。PKU-MMD包含两个阶段，用于难度越来越大的动作检测任务。阶段1是大裕度动作检测任务。阶段2是小余量动作检测任务。我们的数据集还提供多模态数据源，包括RGB、深度、红外辐射和骨架。我们相信，这个大规模数据集将有益于社区行动检测的未来研究。

PKU-MMD is a novel large-scale benchmark for continuous multimodal 3D human action understanding, covering a wide range of complex human activities with well-annotated information. It consists of two stages designed for action detection tasks with escalating difficulty: Stage 1 targets large-margin action detection, and Stage 2 focuses on small-margin action detection. The dataset also provides multimodal data sources including RGB, depth, infrared radiation, and skeletal data. We believe this large-scale benchmark will facilitate future research on action detection in the academic community.

搜集汇总

背景与挑战

背景概述

PKU-MMD是一个大规模多模态3D人类动作理解基准数据集，专注于连续动作检测任务，包含两个难度递增的阶段（大裕度和小余量检测）。它提供RGB、深度、红外辐射和骨架等多模态数据源，适用于复杂人类活动分析，由北京大学和微软亚洲研究院于2017年发布，旨在推动动作检测研究。