The paper introduces a large-scale RGB+D dataset for 3D human activity analysis, named NTU RGB+D, which includes over 56,000 video samples and 4 million frames from 40 distinct subjects. The dataset covers 60 different action classes, including daily, mutual, and health-related actions. To address the limitations of existing datasets, such as small sample sizes, limited camera views, and narrow age ranges, the NTU RGB+D dataset offers a more diverse and comprehensive collection of data. The authors also propose a novel part-aware long short-term memory (LSTM) network to model the long-term temporal correlations of body parts, enhancing the performance of action classification. Experimental results demonstrate the superiority of deep learning methods over traditional hand-crafted features on cross-subject and cross-view evaluation criteria. The dataset and proposed network are expected to advance the field of 3D human activity analysis by enabling the application of data-hungry learning techniques.The paper introduces a large-scale RGB+D dataset for 3D human activity analysis, named NTU RGB+D, which includes over 56,000 video samples and 4 million frames from 40 distinct subjects. The dataset covers 60 different action classes, including daily, mutual, and health-related actions. To address the limitations of existing datasets, such as small sample sizes, limited camera views, and narrow age ranges, the NTU RGB+D dataset offers a more diverse and comprehensive collection of data. The authors also propose a novel part-aware long short-term memory (LSTM) network to model the long-term temporal correlations of body parts, enhancing the performance of action classification. Experimental results demonstrate the superiority of deep learning methods over traditional hand-crafted features on cross-subject and cross-view evaluation criteria. The dataset and proposed network are expected to advance the field of 3D human activity analysis by enabling the application of data-hungry learning techniques.