alternative

A Large-Scale Dataset for Perception in Unstructured Environments

GOOSE - the German Outdoor and Offroad Dataset - is a modern dataset specification and accompanying off-road dataset which is currently under development. In contrast to existing datasets, the focus is on unstructured off-road environments as well as on a broad support for different platforms and applications in the fields of mobile robotics and deep learning. The data is published under the CC BY-SA license and contains annotated point clouds, segmentation masks from RGB and NIR images, as well as accurate localization.

Download Dataset
ZIP files with ground truth data
Access Database
Search and download sequences with metadata
MuCar-v3
Logos

Timeline

  1. Documentation

    Basic platform documentation, training examples and pretrained weights.

  2. Annotated Data

    ZIP files of training set including annotated RGB images and 3D Pointclouds.

  3. Metadata Database

    Metadata bagfiles containing entire sequences with additional sensor raw data.

  4. Additional Plaforms

    Additional data release from an excavator and a legged robot.

  5. Challenge

    Setup of semantic segmentation challenge leaderboard.

Demo

The sequences were recorderd under different terrain and weather conditions to allow for a variety of applications.

alternative

Features

Rich Annotations

Instance-segmented RGB/NIR images correspond to pointwise annotated pointclouds from multiple LiDARs.

Lots of Metadata

Additional data such as localization, additional camera views and raw sensor data from the whole sequence.

Ready for Robotics

Ready-to-use for robotics development: We provide ROS-bags, code snippets and an extensive documentation.

Consistent Ontology

Well defined onotology to be easily extended or used with other datasets.

Extensible

Anyone can contribute data to the GOOSE database server using a well-defined scheme.

Large Scale

15000 image and point cloud frames in various environments.

1

Annotated Frames

1

Classes

1

Robotic Platforms

1

Hours of Recorded Data

1

Sequences

Documentation

We provide extensive documentation for the dataset, including the labeling policy, code snippets, platform descriptions, calibration data, etc.

Access Documentation

Publications

If you use the GOOSE Datasets in your work, please consider citing the following publications:

GOOSE Publication GOOSE-Ex Publication
Bibtex 
                        
@article{goose-dataset,
    author = {Peter Mortimer and Raphael Hagmanns and Miguel Granero
              and Thorsten Luettel and Janko Petereit and Hans-Joachim Wuensche},
    title = {The GOOSE Dataset for Perception in Unstructured Environments},
    url={https://arxiv.org/abs/2310.16788},
    conference={2024 IEEE International Conference on Robotics and Automation (ICRA)}
    year = 2024
}

@article{goose-ex-dataset,
    author = {Raphael Hagmanns and Peter Mortimer and Miguel Granero
              and Thorsten Luettel and Janko Petereit},
    title = {Excavating in the Wild: The GOOSE-Ex Dataset for Semantic Segmentation},
    url={},
    conference={TBA}
    year = 2024
}

Contributors

GOOSE is a project of Fraunhofer IOSB, UniBW Munich and University of Koblenz.

Contributor 1

Raphael Hagmanns

Fraunhofer IOSB

Contributor 2

Peter Mortimer

UniBW Munich

Contributor 3

Miguel Granero

Fraunhofer IOSB

Contributor 4

Roman Abayev

University of Koblenz

Contributor 5

Janko Petereit

Fraunhofer IOSB

Contributor 6

Anselm von Gladiß

University of Koblenz

Contributor 7

Thorsten Lüttel

UniBW Munich

Copyright © Fraunhofer IOSB