Campus news in brief
New coding method allows computer systems to better identify faces in crowds
Deva Ramanan, associate professor of robotics at Carnegie Mellon, and Peiyun Hu, a Ph.D. candidate in robotics, have made significant progress to allow computer vision systems to be able to detect tiny faces. This is due to an improved coding method that allows this system to look for larger things associated with tiny objects.
Of the faces analyzed using Ramanan and Hu’s method, 81 percent proved to be actual faces. This is compared to the 29 to 64 percent found using prior methods.
Ramanan describes this accuracy as being similar to spotting a toothpick in someone's hand. “The toothpick is easier to see when you have hints that someone might be using a toothpick," said Ramanan in a university press release. "For that, the orientation of the fingers and the motion and position of the hand are major clues.”
Similarly, to find a face that may be only a few pixels in size, it helps to first look for a body within the larger image, or to realize that an image also contains a crowd of people.
Ramanan and Hu's method uses "foveal descriptors" which allow them to encode context similar to how human vision is structured. The center of the human field of vision is focused on the retina’s fovea, where visual acuity is highest. The foveal descriptor provides sharp detail for a small patch of the image, while blurring the peripheral image to understand the patch shown in high focus while not overwhelming the computer.
The researchers will present their findings at the Computer Vision and Pattern Recognition conference from July 21-26 in Honolulu.
**Carnegie Mellon students compete in this year’s Three Minute Thesis competition **
Thus far, nine students have been identified as finalists for Carnegie Mellon's Three Minute Thesis competition which challenges Ph.D. students to present a compelling and easily understood presentation of their thesis and its significance in just three minutes.
This is Carnegie Mellon University's fourth year hosting the event, which started at the University of Queensland in 2008. Keith Webster, Dean of University Libraries, brought the concept to Carnegie Mellon and now hosts the competition.
"Our students are doing such interesting, innovative, and complex work," Webster said in a university press release. "It's a joy to learn more about their research and see how they approach the challenge of conveying it to a non-specialist audience."
This year, 78 graduate students participated in 10 rounds of preliminary competitions, which has been the university's highest number of participants in the competition.
Michael Craig, a doctoral student in the College of Engineering, will be participating in the program for his third time this year. His talk for this year is titled "Grid-scale electricity storage: A help or hindrance for mitigating climate change?"
"Last year I got too involved in the details," Craig said in a university press release. "But it's such a good experience. More people should participate."
Ania Jaroszewicz, who is pursuing a doctoral degree in behavioral decision research in Dietrich College's Department of Social and Decision Sciences, is competing this year for the first time. She says that Three Minute Thesis has given her an opportunity to think about how she communicates her work to people who might use it to help shape policy or nonprofit programming.
A panel of judges will select the overall winner and runner-up for the competition.