The recent advances in computer vision have been mainly driven by deep learning, with no end in sight. Autonomous driving is a very exciting application domain for a lot of computer vision problems. In a vehicle, energy resources are very limited. At the same time, there is a trend towards more sensors with higher resolutions. This exposes a contradiction in autonomous driving: huge amounts of data have to be processed as fast, accurate and with as little power consumption as possible.

This workshop will encourage work on efficient neural networks in the context of autonomous driving. We challenge the IV community to participate in a semantic segmentation challenge, offering a prize for the winner. The workshop program will include invited talks, presentations of the top 3 challenge winners and a poster session of submitted work.

Challenge Details
The prize for the winning team is awarded by a jury. The prize will be given to the team showing the best compromise between high accuracy, high speed in frames/second and low power consumption. The winner team is required to give a detailed presentation (with reproducible results) at the workshop. Papers about the work are optional but highly welcome to a special issue in IEEE T-IV.

The accuracy of the system will be measured following the Cityscapes1,2 benchmark for class segmentation. Testing will be done using unpublished data taken with the same car and camera system that was used for the Cityscapes sequences. As a consequence, you have to process 2048x1024px RGB images and output 2048x1024px maps of labels according to the Cityscapes label set3. Prior to the deadline, a sequence of roughly one minute will be supplied to interested teams. The results (whole sequence) have to be submitted within a day, using the same labels and format as known from the standard Cityscapes benchmark.

For a fair comparison, submitting teams have to measure the speed of their solution and to specify the used HW.

The deadline for submissions is May, 1st. Participants are requested to render a video of the entire sequence and to show the results during their presentation at the workshop.

References:
[1] Cityscapes Dataset: M. Cordts, M. Omran, S. Ramos, T. Rehfeld, M. Enzweiler, R. Benenson, U. Franke, S. Roth, and B. Schiele, "The Cityscapes Dataset for Semantic Urban Scene Understanding," in Proc. of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016. https://arxiv.org/abs/1604.01685
[2] Cityscapes Homepage: https://www.cityscapes-dataset.com/
[3] Cityscapes Github: https://github.com/mcordts/cityscapesScripts
For the labelset see: https://github.com/mcordts/cityscapesScripts/blob/master/cityscapesscripts/helpers/labels.py