A Large-Scale, Evolving Multi-modal Dataset for Environmental Perception Based on Commercial off-the-shelf (COTS) 5G/5G-A gNB Devices

CellularEye is a pioneering large-scale multi-modal dataset designed for cutting-edge environmental perception research. Its core feature is the use of commercial communication equipment (BBU, AAU) to collect real-world cellular network IQ data, synchronized with high-resolution visible-light video, infrared video, and weather data. Our goal is to bridge the gap between communication and sensing, providing robust, real-world data support for researchers exploring the future of Integrated Sensing and Communication (ISAC).

• Commercial Cellular Signals: Data originates from operational, commercial cellular network equipment, not simulations or lab-grade signals, offering high research value.
• Rich Multi-modal Data: Includes tightly synchronized IQ data streams, visible-light video, infrared video, and detailed meteorological metrics.
• Diverse Scenarios: Covers a wide range of real-world scenarios, including different times of day, weather conditions, and target activities.
• Evolving Dataset: CellularEye is a "living" dataset. We are committed to continuous updates, releasing more scenarios and richer annotations in the future.

This dataset is collected by a cell-free distributed base station system, comprising eCPRI interface millimeter-wave (mmWave) and Sub-6GHz IQ data, infrared video, visible-light video, and corresponding meteorological data. We aim to provide high-quality, real-world multi-modal data for the ISAC field, especially for tasks like environmental perception, object detection, and tracking.
To capture diverse environmental characteristics, we perform synchronized data collection at four key time points daily: 00:00, 06:00, 12:00, and 18:00.

All data is archived by collection sequence (i.e., the start time of the collection task). In the dataset's root directory, you will find the following folder structure:

<dataset_root>/
└── 2025_09_27_00_00/  # Collection id (YYYY_MM_DD_HH_MM)
    ├── camera/
    │   ├── camera_2025-09-27-00_01_01.mp4
    │   ├── ir_2025-09-27-00_01_01.mp4
    │   └── ...
    ├── meteorological/
    │   ├── 2025-09-27-00-01-01.txt
    │   └── ...
    └── mmw/
        ├── 21/
        │   ├── 2025_09_27_00_01_01_280.bin
        │   └── ...
        ├── 22/
        ├── 23/
        └── 24/
└── 2025_09_27_06_00/
    └── ...

Each .bin file represents one sensing frame. The data arrangement within the .bin file is shown in the figure below.

The system's sensing parameters are shown in the figure below.

Purple Mountain Laboratories (PML) 6G Verification Center, Outdoor Testbed

For your convenience, we provide a Python Script to help you easily read and visualize the data. You can run the script as follows:

python rv_public_v3.py --bin_dir /public_data/2025_10_19_18_00/mmw --bs_id 22 --beam_id 30

Download Data: We recommend downloading the dataset via the following links. To ensure the reproducibility of your research, please explicitly state the dataset version you used in your paper.

We are delighted to see the CellularEye dataset being used in the following cutting-edge research. If you have also used this dataset and wish to showcase your work here, please feel free to contact us (or submit a Pull Request on our GitHub repository).

CellularEye is initiated by Purple Mountain Laboratories and supported by the following institutions. We are proud to empower cutting-edge research at universities and labs worldwide.

Using our dataset?
If your team is using CellularEye, please let us know! We would be honored to feature your institution here.