Invitation to the Informatik-Kolloquium
Fri, 29 July 2016 - 11:00 h s.t. - Raum B U101, Oet. 67
Kai Yu was a PhD Student under the supervision of Hans-Peter Kriegel and Volker Tresp and received his PhD from the LMU in 2003. Among other positions, he was the director of the Deep Learning Institute at Baidu Research. Last year, he started his own Company, Horizon Robotics. The press described the company´s Goals as "Artificial Intelligence-On-A-Chip" and "Android of Al".
In the first part of his talk he addresses the increasing importance of Al for Industry in China.
The title of the second part of his presentation is:
A High-Performance Convolutional Neural Network Framework for Object Detection with a Dense Prediction Map
Object Detection is an important computer vision problem for a large variety of applications. In camera-based sensing systems für intelligent vehicles, object detection provides the fundamental ability to real-time environment perception. Recently, convolutional neural networks (CNNs) have led to significant improvements in this area. However, the current state-of-the-art CNN-based detection methods (e.g. R-CNNs) still suffer from low efficiency and pool accuracy on car detection Tasks, where objects usually have big variations in scale and are heavily occluded. In this talk, I will present a fully end-to-end object detection framework, which archives the world best performance on KITTI car detection task. Our Network can run at real-time in KITTI Images, and outperform state-of-the-art methods such a faster R-CNN using VGG16.