The “NTU RGB+D” repository provides access to a large-scale dataset for human action recognition (and its extension, NTU RGB+D 120). The dataset includes multiple modalities (RGB video, depth sequences, infrared video, 3D skeletal joint data) captured with multiple Kinect v2 cameras simultaneously. The repository also contains MATLAB / Python demo scripts for loading, visualizing, and processing skeleton data, mapping between modalities, and handling dataset structure. Multi-modal action recognition dataset, RGB, depth, infrared, skeletal data. Split into background / evaluation sets for one-shot evaluation (in the extended dataset).

Features

  • Multi-modal action recognition dataset: RGB, depth, infrared, skeletal data
  • Large scale: ~56,880 samples in NTU RGB+D, ~114,480 in NTU RGB+D 120
  • Demo / sample code in MATLAB / Python for loading, visualization, preprocessing
  • Standardized naming and folder structure (SsssCcccPpppRrrrAaaa)
  • Some samples with missing or incomplete skeleton data, with lists provided
  • Split into background / evaluation sets for one-shot evaluation (in the extended dataset)

Project Samples

NTU RGB-D Screenshot 1

Project Activity

See All Activity >