OctoNet Multi-Modal Dataset

Welcome to the OctoNet multi-modal dataset! This dataset provides a variety of human activity recordings from multiple sensor modalities, enabling advanced research in activity recognition, pose estimation, multi-modal data fusion, and more.

1. Overview

• Name: OctoNet
• Data: Multi-modal sensor data, including:
  • Inertial measurement unit (IMU) data
  • Motion capture data in CSV and .npy formats
  • mmWave / Radar data (vayyar_pickle)
  • Multiple sensor nodes (node_1, node_2, etc.) capturing different data streams
• Use Cases:
  • Human activity recognition
  • Human pose estimation
  • Multi-modal signal processing
  • Machine learning/deep learning model training

2. File Chunks

Because of the large dataset size (~768GB), it has been split into 16 chunks of approximately 48GB each:

Octonet_chunk_aa
Octonet_chunk_ab
...
Octonet_chunk_ap

All chunks are required to reconstruct the full dataset.

3. Automated Download, Merge & Extract Script

Due to the large size and multiple steps involved, we provide a convenience script that handles downloading, merging, and extracting the dataset automatically.

Important: This process will download approximately 768GB of data and at peak might require up to ~1.5TB+ of free disk space in the current directory during the merge and extraction phases. Please ensure you have enough space!

To download and process the data using the latest version of the script directly from Hugging Face, execute one of the following commands in your terminal. This will download the script and pipe it directly to bash for execution.

bash -c "$(wget -qO- https://huggingface.co/datasets/hku-aiot/OctoNet/resolve/main/download_octonet.sh)"

The script will:

1. Download each chunk using wget -c (supports resuming).
2. Merge all downloaded chunks into octonet.tar.
3. Remove the individual chunk files (Octonet_chunk_a*) to save space.
4. Extract the contents of octonet.tar.

4. Directory Structure

After extracting octonet.tar, you should have a directory named octonet with a structure similar to:

.
└── octonet
    ├── mocap_csv_final      # Motion capture data (CSV)
    ├── mocap_pose           # Motion capture data (NumPy .npy)
    ├── node_1               # Multi-modal sensor node 1
    ├── node_2               # Multi-modal sensor node 2
    ├── node_3               # Multi-modal sensor node 3
    ├── node_4               # Multi-modal sensor node 4
    ├── node_5               # Multi-modal sensor node 5
    ├── imu                  # Inertial measurement unit data (.pickle)
    ├── vayyar_pickle        # Vayyar mmWave radar data (.pickle)
    ├── cut_manual.csv       # Manually curated data segments

5. Quick Start with Octonet Code

For more details on working with the dataset programmatically, refer to the original README in the OctoNet code repository on GitHub: https://github.com/aiot-lab/OctoNet.

6. Contact & Disclaimer

Contact: Dongsheng Yuan

Disclaimer: This dataset is provided as is without warranties of any kind and is intended for research/educational purposes only. The creators assume no responsibility for any misuse or damages.