DeepDriving
Learning Representations
for Intelligent Vehicles

Vision is a rich and un-intrusive sensor modality that can be captured by cheap and flexible sensors. However, its strengths are also its downside as it is challenging to extract what is relevant from the high dimensional signal. Recently computer vision has experienced what can only be referred to as a revolution. The second coming of neural networks, known as deep learning, has lead to a significant increase in performance in tasks across the field. It is now possible to learn image representations directly from data rather than relying on ad-hoc handcrafted features.

The aim of this workshop is to bring together researchers to discuss both theoretical and practical issues related to the application of deep vision techniques for intelligent vehicles. We want to demonstrate what computer vision currently is capable of and identify important directions of future work. The workshop will be centered around a set of invited talks from prominent researchers together with a poster session of submitted extended abstracts as well as a tutorial in the afternoon. The tutorial will focus on the software library caffe, one of the most popular convolutional neural network toolboxes for vision tasks.

We invite the submissions of extended abstracts for poster contributions to the workshop. Topics of interest are:

• Deep learning for reinforcement learning
• Detection and segmentation networks
• Uncertainty propagation in deep neural networks
• Efficient inference with deep neural networks
• Deep neural networks for perception systems in intelligent vehicles
• Datasets for autonomous driving

• Deadline for extended abstract submissions: 15th of April, 2016
• Acceptance notification: 30th of April, 2016
• Workshop: 19th of June, 2016

Extended abstracts need to be submitted by email to Carl Henrik Ek.
We will not use the PaperCept system.

Formatting guidelines and submission rules:

• 1. Maximum of four pages including references
• 2. IEEE format
• 3. The extended abstracts will not be published in the proceedings. However they will be published on this website if the authors agree.
• 4. The submitted abstract should present recent results, but not necessarily novel ones.
• 5. All accepted abstracts will be presented as posters at the workshop. We will select a single abstract which will be additionally given the opportunity for an oral presentation.
• 6. All abstracts will be reviewed by the organizers.

• Trevor Darrell (EECS UC Berkeley, United States)
• Carl Henrik Ek (University of Bristol, United Kingdom)
• Erik Rodner (Computer Vision Group, University of Jena, Germany)
• Jose M. Alvarez (Smart Vision Systems, NICTA, Australia)
• Lars Petersson (Smart Vision Systems, NICTA, Australia)
• Uwe Franke (Daimler AG, Germany)