Visteon’s computer vision-based lane detection challenge – named ViLD – is open to automotive, academic and other deep learning developers and practitioners. Using Visteon’s DriveCore™ Studio software development kit, ViLD offers participants the opportunity to interact and test deep learning solutions directly on the DriveCore™ platform, with the objective of bringing new deep learning research closer to real-world, state-of-the-art automotive grade software.
ViLD is a computer vision-based lane detection challenge which focuses on a very specific task: The detection of lanes painted on the road through a front-mounted camera under a variety of conditions including brightness, color and curvature. Additionally, participants are asked to explore the real-world challenges of such a detection task. For example, instead of focusing on metrics for the quality of detections, we ask how deployable these detectors are. Can they run on edge devices that might already be running many processes? And, how dependable are they? Would they perform well even when the environment changes slightly? For ease of evaluation, it is assumed that these solutions involve deep learning-based approaches.
The challenge is open to everyone around the globe and comes with the opportunity to test the algorithm on DriveCore™, while DriveCore™ Studio can be used to view these metrics. It is associated with the CVPR 2019 workshop DThree that Visteon is organizing in partnership with the University of Oxford, University of Genova and the BMW Group.
Running from March 1-31, participants are invited to download a baseline solution and associated competition files from the ViLD online portal, which will help participants make a quick start. Submissions are made via the portal where a panel of Visteon judges, supported by partners from Oxford and Genova Universities and BMW, will evaluate submissions on a range of qualitative and quantitative criteria over two phases.