Non-commercial research use only. The dataset is licensed under CC BY-NC 4.0. It may be used solely for academic and non-commercial research purposes. Any commercial use requires separate written permission from the authors.
No redistribution. You will not redistribute, publish, or otherwise make the dataset (in whole or in part, in original or derived form) available to third parties. Direct interested parties to this repository instead.
Attribution. Any publication, presentation, or public artifact that uses RadarEyes must cite the DREAM-PCD paper (see the citation section of the dataset card).
No warranty. The dataset is provided "as is", without warranty of any kind. The authors accept no liability for any consequence arising from its use.
Accurate information. The information you provide below is truthful, and you are authorized by your institution to accept these terms on its behalf.

RadarEyes

RadarEyes is a large-scale indoor mmWave radar point cloud dataset with synchronized horizontal radar, vertical radar, LiDAR, and tracking camera. It is the dataset behind the paper DREAM-PCD: Deep Reconstruction and Enhancement of mmWave Radar Pointcloud (IEEE TIP 2024).

Unlike most mmWave datasets, RadarEyes ships raw ADC data together with the loading and processing code, which makes it usable for signal-processing research rather than only for point-cloud-level tasks.

📄 Paper: arXiv:2309.15374 · IEEE TIP
💻 Code: github.com/ruixv/RadarEyes

Sensor platform

Four sensors mounted on a remotely controllable vehicle:

	Horizontal radar	Vertical radar	LiDAR
Type	FMCW	FMCW	Pulse
Frame rate (Hz)	10	100	10
Frequency	77.70–78.90 GHz	79.10–80.70 GHz	193 THz
TX / RX	3 / 4	3 / 4	—
Range resolution (m)	0.12	0.09375	—
Max range (m)	15.99	12.0	—
Azimuth FoV	±50° @6dB	±20° @6dB	±60°
Elevation FoV	±20° @6dB	±50° @6dB	±12.5°

Radars: 2× TI AWR1843BOOST-EVM + DCA1000-EVM (raw ADC capture)
LiDAR: LS-128 S2, 128-beam 1550 nm automotive-grade hybrid solid-state
Camera / IMU: ZED 2i, 30 fps, <1% closed-loop drift

What is in this repository

This repository hosts 6 sequences (~22 GB uncompressed). Per-frame binary directories are shipped as .tar.gz archives — the Hugging Face Hub caps a single folder at 10,000 files, and one sequence exceeds that on its own. Small metadata files (timestamp.txt, pose.txt, …) and the raw UDP captures are stored uncompressed so they stay browsable.

<sequence>/
├── 1843_azi/                   horizontal radar
│   ├── ADC.tar.gz              raw ADC, frame_*.bin
│   ├── PCD.tar.gz              point clouds
│   ├── PCD_SamePaddingUDPERROR.tar.gz
│   ├── PCD_nomultipath.tar.gz
│   ├── mmPCD_filtered_normalthr.tar.gz
│   ├── udpData.dat             raw UDP capture
│   ├── timestamp.txt
│   └── del_frame.txt           frames dropped during parsing
├── 1843_ele/                   vertical radar (same layout)
├── Camera/                     pose.txt, timestamp.txt, velocity.txt
├── Camera_ZED/                 pose.txt, timestamp.txt
├── Lidar/                      frames.tar.gz (frame_*.bin) + timestamp.txt
└── Lidar_pcd/                  frames.tar.gz
Code/                           loading, fusion and visualization scripts

Sequences

Sequence	Scene	Size
`2023_05_18_17_22_18_cars_30s_1`	cars, 30 s	4.3 GB
`2023_05_18_17_23_42_cars_30s_2`	cars, 30 s	2.5 GB
`2023_05_18_17_36_45_4f_401to409`	4th floor, rooms 401→409	8.2 GB
`2023_05_18_17_40_56_408_30s`	room 408, 30 s	2.6 GB
`2023_05_18_17_41_53_408_20s`	room 408, 20 s	1.9 GB
`2023_06_04_08_15_26_B409_1`	room B409	2.0 GB

Note: parsed vertical-radar ADC (1843_ele/ADC.tar.gz) is currently available for 2023_05_18_17_22_18_cars_30s_1 only. For the other sequences the vertical radar is provided as the raw udpData.dat capture, which can be parsed with the scripts in Code/IPLab_mmwavePCD/ParseData/.

Getting started

Access is gated — request access first, then authenticate:

hf auth login

Download a single sequence (recommended for a first look):

hf download ruixu1/RadarEyes --repo-type dataset \
    --include "2023_05_18_17_22_18_cars_30s_1/*" "Code/*" \
    --local-dir ./RadarEyes

Or the whole dataset:

hf download ruixu1/RadarEyes --repo-type dataset --local-dir ./RadarEyes

Unpack the archives in place — this restores the original directory layout expected by the processing code:

cd RadarEyes
find . -name '*.tar.gz' | while read f; do
    d="${f%.tar.gz}"; mkdir -p "$d" && tar -xzf "$f" -C "$d" && rm "$f"
done

Lidar/frames.tar.gz and Lidar_pcd/frames.tar.gz unpack into a frames/ subdirectory; move their contents up one level if you want the exact original layout:

for d in */Lidar */Lidar_pcd; do
    [ -d "$d/frames" ] && mv "$d/frames/"* "$d/" && rmdir "$d/frames"
done

Then follow Code/readme.md:

python Code/IPLab_mmwavePCD/FuseData/new_azi_radar_fuse.py   # fuse horizontal radar
python Code/IPLab_mmwavePCD/FuseData/new_lidar_fuse.py       # fuse LiDAR

License and terms

Released under CC BY-NC 4.0 — attribution required, non-commercial use only. Redistribution is not permitted; please point others to this repository. Commercial use requires separate written permission from the authors.

Citation

@article{geng2024dreampcd,
  title   = {DREAM-PCD: Deep Reconstruction and Enhancement of mmWave Radar Pointcloud},
  author  = {Geng, Ruixu and Li, Yadong and Zhang, Dongheng and Wu, Jincheng
             and Gao, Yating and Hu, Yang and Chen, Yan},
  journal = {IEEE Transactions on Image Processing},
  year    = {2024},
  doi     = {10.1109/TIP.2024.3512356}
}

Contact

Ruixu Geng — gengruixu@mail.ustc.edu.cn

Changelog

2024-12-23 — point clouds re-saved with corrected dimensions. If you downloaded before this date, please overwrite with the current data.

Downloads last month: 14

Total file size:

13 GB

Paper for ruixu1/RadarEyes

DREAM-PCD: Deep Reconstruction and Enhancement of mmWave Radar Pointcloud

Paper • 2309.15374 • Published Sep 27, 2023