CEAS Faculty and Students Win Award in First IEEE AI City Challenge

CVML Team 
The CVML Team: Jonathan Song, Siwei Lyu, Ming-Ching Chang, and Yi Wei; Lipeng Ke not pictured.

A team led by Professors Ming-Ching Chang and Siwei Lyu from UAlbany’s Computer Vision and Machine Learning (CVML) Lab was awarded an "Honorary Mention" in the first IEEE Smart World NVIDIA AI City Challenge on August 5, 2017 in San Jose, California.

The purpose of the challenge was to solve real-world transportation problems related to safety and congestion. This work is important and timely as millions of people are either killed or injured in automobile accidents yearly across the world, and more accurate object detection and classification would advance artificial intelligence solutions in this area. Using authentic traffic camera video data, 28 expert teams from 18 universities collaborated on projects devised to make transportation systems safer and smarter. Participants tested their algorithms to detect meaningful objects, such as vehicles and pedestrians, and automatically analyze traffic flow based on these detections.  

The CVML team received an “Honorary Mention” award for their overall contributions to the competition. Their model in the Vehicle Detection Challenge was ranked number one in best performance in the AIC 1080 test set, and their vehicle detection and tracking results were used as a baseline to support all the other teams in the Smart Transportation Challenge. 

Traffic Analysis Results: Visualizing the moving direction and velocity (MPH) for each vehicle, with a top-down view from Google Map on the right. Enlarge photo. (Also view other selected model images.)

"The technique that the UAlbany team used is a novel method based on deep learning, which solves complicated AI problems with neural networks with many layers of artificial neurons,” said Lyu.  

Lyu is alluding to the artificial neural networks that are used in machine learning. Think of the brain: artificial neural networks are akin to biological neural networks such as the human brain, in that they function as a system of interconnected artificial neurons that communicate with each other.  

“Deep learning is the same technique behind the recent success of Google AlphaGo and Tesla's self-driving car. Our team has been working at the cutting edge of this exciting technology, and the algorithm we developed is also applicable to many other problems that require automatic and accurate detections, for instance, specific cell types in medical imaging and abnormal patterns in network communication,”​ Lyu said.

Lyu of Computer Science, and Chang of Electrical and Computer Engineering, collaborated on the development of their model with Computer Science doctoral students Yi Wei and Jonathan Song, and Lipeng Ke, a visiting student from the University of Chinese Academy of Sciences in Beijing.

The competition experience was very rewarding for the three graduate students on the team. Wei, who learned how to train and tune a deep neural network to detect objects, says, “It was a great experience for me; applying state-of-the-art artificial intelligence algorithms to solve real world applications brings me a sense of achievement.”  

Song, who also trained the object detection model by localizing and identifying multiple objects in a single image, says, “It was amazing experiencing firsthand how creativity and state-of-the-art technology can advance human life---- by applying artificial intelligence and computer vision on an autonomous car to make our travel safer and more efficient!”  

Lipeng Ke

Lipeng Ke

Ke said, “It is a great honor to apply our research to such a competition organized by industry and academia and to be recognized for our achievement in research. I am very proud of our team's achievement and very excited about the broad use of the application of our research.”  

CEAS Dean Kim Boyer said, “We are extremely proud of the accomplishments of Professors Lyu and Chang and their group. Humans are effectively ‘Mozarts of Vision,’ meaning that we find seeing to be as effortless as he found composing music.  So most of us fail to appreciate just how difficult it is to build machines that can see – that is, interpret and assign semantic meaning to the objects and actions in a scene.  I’ve worked on that problem for more than 30 years – and I can tell you theirs is a noteworthy accomplishment.”  

The competition was jointly hosted by IEEE, the world’s largest and most prestigious professional organization for electrical and computer engineers and related sciences, and NVIDIA, a technology company based in Portland, Oregon, originally specializing in making graphical processing units (GPUs) for the video game industry, but which in recent years has developed GPU processors for deep learning. Chang is a member of IEEE. Lyu has been an IEEE Senior Member since 2016, and received the National Science Foundation’s prestigious CAREER Award in 2011.  

Established in 2009, the CVML Lab is a joint lab of the Computer Science and Electrical and Computer Engineering Departments of the College of Engineering and Applied Sciences. It is the spearhead research group in AI, computer vision, and machine learning at the University. Funding for the lab has been provided by the National Science Foundation (NSF), the Defense Advance Program Research Agency (DARPA), and the National Institute of Justice (NIJ), and generous donations from NVidia Inc. Lyu and Chang are the Director and co-Director of CVML, respectively.   For more about Professors Lyu and Chang and their research, please visit their faculty pages, as well as the web page for CVML

Written by Daphne Jorgensen