A Fine-Grained Dataset and its Efficient Semantic
Segmentation for Unstructured Driving Scenarios

Kai A. Metzger

Peter Mortimer

Hans-Joachim Wuensche

Institute for Autonomous Systems Technology

Bundeswehr University Munich

In ICPR, 2020

Overview. Example images from the TAS500 dataset.
The TAS500 dataset contains fine-grained terrain and vegetation annotations
of over 500 scenes in unstructured environments.

Research in autonomous driving for unstructured environments suffers from a lack of semantically labeled datasets compared to its urban counterpart. Urban and unstructured outdoor environments are challenging due to the varying lighting and weather conditions during a day and across seasons. In this paper, we introduce TAS500, a novel semantic segmentation dataset for autonomous driving in unstructured environments. TAS500 offers fine-grained vegetation and terrain classes to learn drivable surfaces and natural obstacles in outdoor scenes effectively. We evaluate the performance of modern semantic segmentation models with an additional focus on their efficiency. Our experiments demonstrate the advantages of fine-grained semantic classes to improve the overall prediction accuracy, especially along the class boundaries.

Announcements

2021-01-10: Initial release of the TAS500v1.0 dataset as part of the ICPR publication
2021-04-15: Release of the TAS500v1.1 dataset. In the updated dataset we polished the semantic maps and cropped all driving images to the fixed image size of 620px x 2026px. Download the TAS500v1.1 dataset here.
2021-04-27^New!: The Outdoor Semantic Segmentation Challenge is now available on CodaLab. The task is to develop an algorithm for a fine-grained vegetation and surface detection in an unstructured outdoor environment and the training is done on the TAS500v1.1 dataset. The best results in the competition are presented at the SUUE workshop at the DAGM GCPR 2021.

Paper

Metzger, Mortimer, Wuensche.

A Fine-Grained Dataset and its Efficient Semantic
Segmentation for Unstructured Driving Scenarios

ICPR, 2020.

[pdf]

[Bibtex]

Dataset

Class distribution in the TAS500 dataset. Number of fine-grained pixels (y-axis) per class and their associated category (x-axis).

Data collection pipeline. The data was collected using the autonomous vehicle MuCAR-3. We recorded data with a frame rate of 10 Hz and cut off most of the sky and ego vehicle hood from all images. The final images have a resolution of 620px x 2026px. Our label rate amounts to around 0.1 Hz, and we consequently provide a pixel-wise semantic mask for every hundredth recorded image.

Download

The TAS500 dataset is copyrighted and published under the Attribution-NonCommercial-ShareAlike 3.0 license. This means that you must attribute the work in the manner specified by the authors, you may not use this work for commercial purposes and if you alter, transform, or build upon this work, you may distribute the resulting work only under the same license.

Acknowledgements

The authors gratefully acknowledge funding by the Federal Office of Bundeswehr Equipment, Information Technology and In-Service Support (BAAINBw). This webpage template is based on the Factored3D project page.