Datasets:
The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.
DurLAR: A High-Fidelity 128-Channel LiDAR Dataset
News
- [2025/03/16] We provide the dataset download link hosted on Hugging Face, along with the corresponding new download script.
- [2025/03/16] We provide an exemplar dataset (600 frames), which is now available for direct preview on the webpage, displaying dataset image-based topics (image_00, image_01, reflectivity, ambient).
- [2024/12/05] We provide the intrinsic parameters of our OS1-128 LiDAR [download].
Sensor placement
LiDAR: Ouster OS1-128 LiDAR sensor with 128 channels vertical resolution
Stereo Camera: Carnegie Robotics MultiSense S21 stereo camera with grayscale, colour, and IR enhanced imagers, 2048x1088 @ 2MP resolution
GNSS/INS: OxTS RT3000v3 global navigation satellite and inertial navigation system, supporting localization from GPS, GLONASS, BeiDou, Galileo, PPP and SBAS constellations
Lux Meter: Yocto Light V3, a USB ambient light sensor (lux meter), measuring ambient light up to 100,000 lux
Panoramic Imagery
Reflectivity imagery
Ambient imagery
File Description
Each file contains 8 topics for each frame in DurLAR dataset,
ambient/
: panoramic ambient imageryreflec/
: panoramic reflectivity imageryimage_01/
: right camera (grayscale+synced+rectified)image_02/
: left RGB camera (synced+rectified)ouster_points
: ouster LiDAR point cloud (KITTI-compatible binary format)gps
,imu
,lux
: csv file format
The structure of the provided DurLAR full dataset zip file,
DurLAR_<date>/
โโโ ambient/
โ โโโ data/
โ โ โโโ <frame_number.png> [ ..... ]
โ โโโ timestamp.txt
โโโ gps/
โ โโโ data.csv
โโโ image_01/
โ โโโ data/
โ โ โโโ <frame_number.png> [ ..... ]
โ โโโ timestamp.txt
โโโ image_02/
โ โโโ data/
โ โ โโโ <frame_number.png> [ ..... ]
โ โโโ timestamp.txt
โโโ imu/
โ โโโ data.csv
โโโ lux/
โ โโโ data.csv
โโโ ouster_points/
โ โโโ data/
โ โ โโโ <frame_number.bin> [ ..... ]
โ โโโ timestamp.txt
โโโ reflec/
โ โโโ data/
โ โ โโโ <frame_number.png> [ ..... ]
โ โโโ timestamp.txt
โโโ readme.md [ this README file ]
The structure of the provided calibration zip file,
DurLAR_calibs/
โโโ calib_cam_to_cam.txt [ Camera to camera calibration results ]
โโโ calib_imu_to_lidar.txt [ IMU to LiDAR calibration results ]
โโโ calib_lidar_to_cam.txt [ LiDAR to camera calibration results ]
Get Started
- A quick look: the exemplar dataset (600 frames)
- The full dataset hosted on Hugging Face
- Download the calibration files
- Download the calibration files (v2, targetless)
- Download the exemplar ROS bag (for targetless calibration)
- Download the exemplar dataset (600 frames)
- Download the full dataset (Fill in the form to request access to the full dataset)
Note that we did not include CSV header information in the exemplar dataset (600 frames). You can refer to Header of csv files to get the first line of the
csv
files.
calibration files (v2, targetless): Following the publication of the proposed DurLAR dataset and the corresponding paper, we identify a more advanced targetless calibration method (#4) that surpasses the LiDAR-camera calibration technique previously employed. We provide exemplar ROS bag for targetless calibration, and also corresponding calibration results (v2). Please refer to Appendix (arXiv) for more details.
Prerequisites
Before starting, ensure you have the Hugging Face CLI (huggingface_hub) installed:
โ Install it via:
pip install -U "huggingface_hub[cli]"
Note: The Hugging Face service may be restricted in certain countries or regions. In such cases, you may consider using a Hugging Face mirror (Hugging Face ้ๅ) as an alternative.
Download the dataset using scripts (recommended)
We provide a pre-written Bash script to download the dataset from Hugging Face. You need to manually modify the User Configuration (Modify as Needed) section at the beginning of durlar_hf_download_script.sh to match your desired paths and features.
If you encounter any issues (e.g., network problems or unexpected interruptions), you can also modify this script to fit your needs. For example, if the extraction process is interrupted, you can manually comment out the dataset download section and resume from extraction.
Manually download the dataset (fallback)
If the above script does not work, you can download the dataset manually. This option is only a fallback in case the script fails and is not recommended.
To download all dataset files from Hugging Face, use:
huggingface-cli download l1997i/DurLAR --repo-type dataset --local-dir [YOUR_DATASET_DOWNLOAD_PATH]
The parameter --local-dir
is optional. For example,
# Example 1: The recommended (and default) way to download files from the Hub is to use the cache-system, without --local-dir specified
huggingface-cli download l1997i/DurLAR --repo-type dataset
# Example 2: in some cases you want to download files and move them to a specific folder. You can do that using the --local-dir option.
huggingface-cli download l1997i/DurLAR --repo-type dataset --local-dir /home/my_username/datasets/DurLAR_tar/
Merge and Extract the Dataset
Reassemble the .tar parts:
cd /home/my_username/datasets/DurLAR_tar/ # Enter the dataset local-dir folder. Note: Your path may be different from mine. Please adjust it accordingly.
cat DurLAR_dataset.tar* > DurLAR_dataset_full.tar
Extract the full dataset:
tar -xvf DurLAR_dataset_full.tar -C /your/desired/path
Optionally, If you have pigz installed, you can extract using multi-threaded pigz for faster speed:
which pigz # it should correctly return the path of pigz
tar --use-compress-program="pigz -p $(nproc)" -xvf DurLAR_dataset/DurLAR_dataset.tar -C /your/desired/path
Cleanup (Optional)
Once extracted, delete the archive parts to free up space:
rm -rf DurLAR_tar # Note: Your path may be different from mine. Please adjust it accordingly.
Now you will find the fully extracted DurLAR dataset in your /your/desired/path
.
[Outdated] Early download options for the dataset (not recommended)
The following methods were the early download options for the DurLAR dataset. While they are still functional, they may suffer from drawbacks such as slow download speeds (the host server is in the UK), temporary connection bans, and lack of download resumption. Therefore, we do not recommend using these methods anymore.
Access to the complete DurLAR dataset can be requested through one of the following ways. ๆจๅฏไปป้ไปฅไธๅ ถไธญไปปๆ้พๆฅ็ณ่ฏท่ฎฟ้ฎๅฎๆดๆฐๆฎ้ใ
1. Access for the full dataset
2. ็ณ่ฏท่ฎฟ้ฎๅฎๆดๆฐๆฎ้
[Outdated] Usage of the download script (not recommended)
Upon completion of the form, the download script durlar_download
and accompanying instructions will be automatically provided. The DurLAR dataset can then be downloaded via the command line.
For the first use, it is highly likely that the durlar_download
file will need to be made
executable:
chmod +x durlar_download
By default, this script downloads the small subset for simple testing. Use the following command:
./durlar_download
It is also possible to select and download various test drives:
usage: ./durlar_download [dataset_sample_size] [drive]
dataset_sample_size = [ small | medium | full ]
drive = 1 ... 5
Given the substantial size of the DurLAR dataset, please download the complete dataset only when necessary:
./durlar_download full 5
Throughout the entire download process, it is important that your network remains stable and free from any interruptions. In the event of network issues, please delete all DurLAR dataset folders and rerun the download script. Currently, our script supports only Ubuntu (tested on Ubuntu 18.04 and Ubuntu 20.04, amd64). For downloading the DurLAR dataset on other operating systems, please refer to Durham Collections for instructions.
CSV format for imu
, gps
, and lux
topics
Format description
Our imu
, gps
, and lux
data are all in CSV
format. The first row of the CSV
file contains headers that describe the meaning of each column. Taking imu
csv file for example (only the first 9 rows are displayed),
%time
: Timestamps in Unix epoch format.field.header.seq
: Sequence numbers.field.header.stamp
: Header timestamps.field.header.frame_id
: Frame of reference, labeled as "gps".field.orientation.x
: X-component of the orientation quaternion.field.orientation.y
: Y-component of the orientation quaternion.field.orientation.z
: Z-component of the orientation quaternion.field.orientation.w
: W-component of the orientation quaternion.field.orientation_covariance0
: Covariance of the orientation data.
Header of csv
files
The first line of the csv
files is shown as follows.
For the GPS,
time,field.header.seq,field.header.stamp,field.header.frame_id,field.status.status,field.status.service,field.latitude,field.longitude,field.altitude,field.position_covariance0,field.position_covariance1,field.position_covariance2,field.position_covariance3,field.position_covariance4,field.position_covariance5,field.position_covariance6,field.position_covariance7,field.position_covariance8,field.position_covariance_type
For the IMU,
time,field.header.seq,field.header.stamp,field.header.frame_id,field.orientation.x,field.orientation.y,field.orientation.z,field.orientation.w,field.orientation_covariance0,field.orientation_covariance1,field.orientation_covariance2,field.orientation_covariance3,field.orientation_covariance4,field.orientation_covariance5,field.orientation_covariance6,field.orientation_covariance7,field.orientation_covariance8,field.angular_velocity.x,field.angular_velocity.y,field.angular_velocity.z,field.angular_velocity_covariance0,field.angular_velocity_covariance1,field.angular_velocity_covariance2,field.angular_velocity_covariance3,field.angular_velocity_covariance4,field.angular_velocity_covariance5,field.angular_velocity_covariance6,field.angular_velocity_covariance7,field.angular_velocity_covariance8,field.linear_acceleration.x,field.linear_acceleration.y,field.linear_acceleration.z,field.linear_acceleration_covariance0,field.linear_acceleration_covariance1,field.linear_acceleration_covariance2,field.linear_acceleration_covariance3,field.linear_acceleration_covariance4,field.linear_acceleration_covariance5,field.linear_acceleration_covariance6,field.linear_acceleration_covariance7,field.linear_acceleration_covariance8
For the LUX,
time,field.header.seq,field.header.stamp,field.header.frame_id,field.illuminance,field.variance
To process the csv
files
To process the csv
files, you can use multiple ways. For example,
Python: Use the pandas library to read the CSV file with the following code:
import pandas as pd
df = pd.read_csv('data.csv')
print(df)
Text Editors: Simple text editors like Notepad
(Windows) or TextEdit
(Mac) can also open CSV
files, though they are less suited for data analysis.
Folder #Frame Verification
For easy verification of folder data and integrity, we provide the number of frames in each drive folder, as well as the MD5 checksums of the zip files.
Folder | # of Frames |
---|---|
20210716 | 41993 |
20210901 | 23347 |
20211012 | 28642 |
20211208 | 26850 |
20211209 | 25079 |
total | 145911 |
Intrinsic Parameters of Our Ouster OS1-128 LiDAR
The intrinsic JSON file of our LiDAR can be downloaded at this link. For more information, visit the official user manual of OS1-128.
Please note that sensitive information, such as the serial number and unique device ID, has been redacted (indicated as XXXXXXX).
Reference
If you are making use of this work in any way (including our dataset and toolkits), you must please reference the following paper in any report, publication, presentation, software release or any other associated materials:
DurLAR: A High-fidelity 128-channel LiDAR Dataset with Panoramic Ambient and Reflectivity Imagery for Multi-modal Autonomous Driving Applications (Li Li, Khalid N. Ismail, Hubert P. H. Shum and Toby P. Breckon), In Int. Conf. 3D Vision, 2021. [pdf] [video][poster]
@inproceedings{li21durlar,
author = {Li, L. and Ismail, K.N. and Shum, H.P.H. and Breckon, T.P.},
title = {DurLAR: A High-fidelity 128-channel LiDAR Dataset with Panoramic Ambient and Reflectivity Imagery for Multi-modal Autonomous Driving Applications},
booktitle = {Proc. Int. Conf. on 3D Vision},
year = {2021},
month = {December},
publisher = {IEEE},
keywords = {autonomous driving, dataset, high-resolution LiDAR, flash LiDAR, ground truth depth, dense depth, monocular depth estimation, stereo vision, 3D},
category = {automotive 3Dvision},
}
- Downloads last month
- 2,156