Resources

Cirrus dataset

Cirrus. A curated dataset with unique long-range LiDAR point clouds and scanning patterns, shared to aid advancements in machine perception and safe self-driving technology.

Overview

To date, light detection and ranging (LiDAR) research has relied on standard range point clouds and uniform scanning patterns. With the release of Cirrus, we provide a non-uniform distribution of LiDAR scanning patterns with emphasis on long range. Cirrus also includes corresponding camera images, uniform scanning patterns and annotations.

We hope this will encourage research in algorithm development for long range LiDAR detection and classification.

Cirrus LiDAR range comparison

A diagram comparing Cirrus with two other autonomous driving datasets. Cirrus reaches 250 meters. Kitti reaches 120 meters. NuScenes reaches 70 meters. This gives Cirrus an advantage up to 5.4 seconds on a 120km/h highway.

A comparison between our Cirrus LiDAR and two other autonomous driving datasets, NuScene and KITTI.

Accept functional cookies to view this content.
Click to view your cookie settings.

“Mackerel sky and mares’ tails make lofty ships carry low sails.“

High-altitude Cirrus clouds have historically helped sailors forecast weather conditions, well in advance.

The technology

The Cirrus dataset contains 6,285 pairs of RGB, LiDAR Gaussian, and LiDAR Uniform frames. Cirrus has been annotated for eight object categories (described below) across the entire 250-meter LiDAR effective range. It includes both high-speed highway and low-speed urban-road scenarios. All images have gone through an anonymization process, blurring faces and license plates to eliminate personally identifiable information.

The following sensors were used to collect the Cirrus dataset.

Sensors	Description
Sensors RGB Camera	Description Resolution of 1920 × 650
Sensors 2 x Luminar Hydra LiDAR Sensors	Description 10Hz, 64 lines per frame, 1550-nm, 250m effective range, > 200 meters range to 10% reflective target (Lambertian), 120° horizontal FOV, 30° vertical FOV.
Sensors 2 x GPS and inertial measurement unit (IMU) device	Description Resolution of 1920 × 650

Outlined car from a top view perspective, viewing LiDAR Gaussian, Camera and LiDAR Uniform angles based on the x, y and z axis.

From a windshield perspective. A road with vehicles and street objects by the sides of the road, all categorised and highlighted with squares in different colors.

Dataset with 3D annotations

Each Dataset file (.zip) below includes a whole dataset of Cirrus. The files includes camera images (.jpg) for reference, LiDAR Gaussian (.xyz), LiDAR Uniform (.xyz) and annotation (.json) for the point clouds as well as a matching file (.txt).

Use of the matching file:

Provided annotations are performed on the Gaussian data.
Image frames and Gaussian frames share the same timestamp for 1-to-1 mapping with the annotation files.
A Uniform frame is matched with a Gaussian frame timestamp by using match_uniform*.txt file, where Column 1 = Gaussian file name and Column 2 = Uniform file name.

Dataset 1
898 scenes
|
1.3 GB
Dataset 2
897 scenes
|
1.4 GB
Dataset 3
898 scenes
|
1.3 GB
Dataset 4
898 scenes
|
1.3 GB
Dataset 5
898 scenes
|
1.5 GB
Dataset 6
898 scenes
|
1.5 GB
Dataset 7
898 scenes
|
1.5 GB

1 Zip file

6285 scenes

Total file size 7.0 GB

Annotations

Cirrus contains the 8 following annotated object categories:

VehicleLarge vehiclePedestrianBicycleAnimalWheeled pedestrianTrailerMotorcycle

Privacy

Volvo Cars takes reasonable care to remove or hide personal data including faces of people and license plates of vehicles.

If you would like us to modify or remove certain images from the Cirrus dataset, please contact developer.portal@volvocars.com.

Download example files

Each of the following .zip files are examples from Cirrus and are taken on a highway. Each example scene includes the following files: camera image (.jpg), LiDAR Gaussian (.xyz), LiDAR Uniform (.xyz) and annotation (.json).

Scene 1
1.3 MB
Scene 2
1.2 MB
Scene 3
1.3 MB
Scene 4
1.4 MB
Scene 5
1.2 MB

1 Zip file

5 scenes

Total file size 5.4 MB

2D annotations

Each file (.zip) below includes 2D annotations of the Cirrus images from respective dataset above, in YOLO format. The annotation files are named exactly as the camera image (.jpg) files. The annotations were created using inference of the YOLOv7 model. Hence, the object categories are different in the 2D annotations than in Cirrus. Read more about the categories in the section below.

2D annotations for dataset 1
897 files
|
747 KB
2D annotations for dataset 2
896 files
|
653 KB
2D annotations for dataset 3
895 files
|
628 KB
2D annotations for dataset 4
892 files
|
616 KB
2D annotations for dataset 5
879 files
|
652 KB
2D annotations for dataset 6
897 files
|
698 KB
2D annotations for dataset 7
897 files
|
697 KB

1 Zip file

6260 files

Total file size 5.5 MB

Annotations

The 2D annotations contains the 8 following annotated object categories:

BicycleBusCarMotorcyclePersonStop signTruckTraffic light

The object category distribution of the given 8 categories.

Details

The 2D annotations are done following the YOLO format where the category IDs are zero-based index representation of all available categories. In the YOLO annotation format, all .jpg files from the Cirrus datasets have their corresponding .txt annotation file where total number of lines in the .txt file is equal to the total number of objects present in the .jpg image.

The annotation format is as follows:

Data	Description
Data <category>	Description Zero-based index representing the object category; 0 = Person 1 = Bicycle 2 = Car 3 = Motorcycle 4 = Bus 5 = Truck 6 = Traffic light 7 = Stop sign
Data <x_center> <y_center> <width> <height>	Description Float values relative to width and height of image. It will be in the range (0.0 to 1.0).
Data <x_center>	Description <absolute_x> / <image_width>
Data <y_center>	Description <absolute_y> / <image_height>
Data <width>	Description <bounding_box_width> / <image_width>
Data <height>	Description <bounding_box_height> / <image_height>

Legal notice

Luminar Technologies, Inc. is the sole and exclusive owner of this dataset.
The dataset is licensed under CC BY-SA 4.0.
The 2D annotations is owned by Volvo Cars under the same license as the Cirrus dataset.
Any public use, distribution, display of this data set must contain this notice in its entirety.

Public distribution

When using the Cirrus dataset for public distribution, we would be glad if you cite us. Please cite the following:

Dataset with 3D annotations usage

@inproceedings{wang2019range,
  title = {Range adaptation for 3d object detection in lidar},
  author = {Wang, Ze and Ding, Sihao and Li, Ying and Zhao, Minming and Roychowdhury, Sohini and Wallin, Andreas and Sapiro, Guillermo and Qiu, Qiang},
  booktitle = {Proceedings of the IEEE International Conference on Computer Vision Workshops},
  year = {2019}
}

@misc{Cirrus_dataset,
  title = {Cirrus: A Long-range Bi-pattern LiDAR Dataset},
  author = {Wang, Ze and Ding, Sihao and Li, Ying and Zhao, Minming and Roychowdhury, Sohini and Wallin, Andreas and Fenn, Jonas and Sapiro, Guillermo and Qiu, Qiang and Martin, Lane and Ryvola, Scott},
  website = {\\url{https://arxiv.org/abs/2012.02938}},
  year = {2020}
}

@misc{Cirrus_dataset,
  title = {Cirrus dataset},
  website = {\\url{https://developer.volvocars.com/resources/cirrus}},
  copyright = {Luminar Technologies, Inc.},
  license = {CC BY-SA 4.0},
  year = {2020}
}

2D annotations usage

@misc{2d_annotations_for_Cirrus_dataset,
  title = {2d annotations},
  website = {\\url{https://developer.volvocars.com/resources/cirrus}},
  copyright = {Volvo Cars Corporation},
  license = {CC BY-SA 4.0},
  year = {2023}
}