Speeding up semantic segmentation for autonomous driving

Michael Treml; José Antonio Arjona Medina; Thomas Unterthiner; Rupesh Durgesh; Felix Friedmann; Peter Schuberth; Andreas Mayr; Martin Heusel; Markus Hofmarcher; Michael Widrich; Ulrich Bodenhofer; Bernhard Nessler; Sepp Hochreiter

Speeding up semantic segmentation for autonomous driving

Michael Treml, José Antonio Arjona Medina, Thomas Unterthiner, Rupesh Durgesh, Felix Friedmann, Peter Schuberth, Andreas Mayr, Martin Heusel, Markus Hofmarcher, Michael Widrich, Ulrich Bodenhofer, Bernhard Nessler, Sepp Hochreiter

Research output: Contribution to conference › Paper › peer-review

Abstract

Deep learning has considerably improved semantic image segmentation. However, its high accuracy is traded against larger computational costs which makes it unsuit-able for embedded devices in self-driving cars. We propose a novel deep network architecture for image segmentation that keeps the high accuracy while being efficient enough for embedded devices. The architecture consists of ELU activation functions, a SqueezeNet-like encoder, followed by parallel dilated convolutions, and a decoder with SharpMask-like refinement modules. On the Cityscapes dataset, the new network achieves higher segmentation accuracy than other networks that are tailored to embedded devices. Simultaneously the frame-rate is still sufficiently high for the deployment in autonomous vehicles.

Original language	English (American)
Publication status	Published - Dec 2016
Externally published	Yes
Event	NIPS Workshop on Machine Learning for Intelligent Transportation Systems - Barcelona, Spain Duration: 8 Dec 2016 → 8 Dec 2016

Workshop

Workshop	NIPS Workshop on Machine Learning for Intelligent Transportation Systems
Country/Territory	Spain
City	Barcelona
Period	08.12.2016 → 08.12.2016

Cite this

@conference{e908694ea4b443ca8f02e9a0f04b7c96,

title = "Speeding up semantic segmentation for autonomous driving",

abstract = "Deep learning has considerably improved semantic image segmentation. However, its high accuracy is traded against larger computational costs which makes it unsuit-able for embedded devices in self-driving cars. We propose a novel deep network architecture for image segmentation that keeps the high accuracy while being efficient enough for embedded devices. The architecture consists of ELU activation functions, a SqueezeNet-like encoder, followed by parallel dilated convolutions, and a decoder with SharpMask-like refinement modules. On the Cityscapes dataset, the new network achieves higher segmentation accuracy than other networks that are tailored to embedded devices. Simultaneously the frame-rate is still sufficiently high for the deployment in autonomous vehicles.",

author = "Michael Treml and {Arjona Medina}, {Jos{\'e} Antonio} and Thomas Unterthiner and Rupesh Durgesh and Felix Friedmann and Peter Schuberth and Andreas Mayr and Martin Heusel and Markus Hofmarcher and Michael Widrich and Ulrich Bodenhofer and Bernhard Nessler and Sepp Hochreiter",

year = "2016",

month = dec,

language = "Englisch (Amerika)",

note = "NIPS Workshop on Machine Learning for Intelligent Transportation Systems ; Conference date: 08-12-2016 Through 08-12-2016",

}

Treml, M, Arjona Medina, JA, Unterthiner, T, Durgesh, R, Friedmann, F, Schuberth, P, Mayr, A, Heusel, M, Hofmarcher, M, Widrich, M, Bodenhofer, U, Nessler, B & Hochreiter, S 2016, 'Speeding up semantic segmentation for autonomous driving', Paper presented at NIPS Workshop on Machine Learning for Intelligent Transportation Systems, Barcelona, Spain, 08.12.2016 - 08.12.2016.

TY - CONF

T1 - Speeding up semantic segmentation for autonomous driving

AU - Treml, Michael

AU - Arjona Medina, José Antonio

AU - Unterthiner, Thomas

AU - Durgesh, Rupesh

AU - Friedmann, Felix

AU - Schuberth, Peter

AU - Mayr, Andreas

AU - Heusel, Martin

AU - Hofmarcher, Markus

AU - Widrich, Michael

AU - Bodenhofer, Ulrich

AU - Nessler, Bernhard

AU - Hochreiter, Sepp

PY - 2016/12

Y1 - 2016/12

N2 - Deep learning has considerably improved semantic image segmentation. However, its high accuracy is traded against larger computational costs which makes it unsuit-able for embedded devices in self-driving cars. We propose a novel deep network architecture for image segmentation that keeps the high accuracy while being efficient enough for embedded devices. The architecture consists of ELU activation functions, a SqueezeNet-like encoder, followed by parallel dilated convolutions, and a decoder with SharpMask-like refinement modules. On the Cityscapes dataset, the new network achieves higher segmentation accuracy than other networks that are tailored to embedded devices. Simultaneously the frame-rate is still sufficiently high for the deployment in autonomous vehicles.

AB - Deep learning has considerably improved semantic image segmentation. However, its high accuracy is traded against larger computational costs which makes it unsuit-able for embedded devices in self-driving cars. We propose a novel deep network architecture for image segmentation that keeps the high accuracy while being efficient enough for embedded devices. The architecture consists of ELU activation functions, a SqueezeNet-like encoder, followed by parallel dilated convolutions, and a decoder with SharpMask-like refinement modules. On the Cityscapes dataset, the new network achieves higher segmentation accuracy than other networks that are tailored to embedded devices. Simultaneously the frame-rate is still sufficiently high for the deployment in autonomous vehicles.

M3 - Papier

T2 - NIPS Workshop on Machine Learning for Intelligent Transportation Systems

Y2 - 8 December 2016 through 8 December 2016

ER -