Chinese researchers have developed a new method for improving how computers "see" and "understand" real-world objects by training their vision systems in virtual reality. Professor Kunfeng Wang at China's State Key Laboratory for Management and Control for Complex Systems says the research team seeks to overcome the drawbacks of real-world image datasets for training and testing vision systems.
The team virtually generated the ParallelEye dataset using commercially available computer software, and modeled an urban area in Beijing by adding various buildings, vehicles, and different weather conditions. Then they positioned a virtual "camera" on a virtual car, which drove around the simulation to produce datasets representative of the real world.
The researchers saw a marked increase in performance on almost every tested metric, and Wang says they also showed that they can easily synthesize diverse sets of images to help build stronger computer vision systems.
From Chinese Association of Automation
View Full Article
Abstracts Copyright © 2018 Information Inc., Bethesda, Maryland, USA
No entries found