The research team led by Prof. Soonhoi Ha, won the Low-Power Image Recognition Challenge 2017 (LIPRC 2017)

- Improved the performance by 2.7 times compared with the last year’s winning team
 

 
Seoul National University (Dean Kunwoo Lee) announced that a research team of Prof. Soonhoi Ha of Dept. of Computer Engineering won the Low-Power Image Recognition Challenge 2017 (LIPRC 2017), which was held in Hawaii on the July 21.
 
The contest was held as part of CVPR 2017 (Computer Vision and Pattern Recognition 2017), the most prestigious academic conference in computer vision and pattern recognition. The goal of this year's competition is to recognize objects in 20,000 images within 10 minutes. The winner was determined by the recognition accuracy per power consumption.
 
The research team was co-directed by Professor Soonhoi Ha and Professor Seungjoo Yoo of the Dept. of Computer Engineering. The team won the competition by a wide margin by adopting the goal-oriented methodologies such as optimal algorithm selection and algorithm parallelization. The competition method is same every year in LPIRC, and the research team improved the performance by 2.7 times compared with the last year's winner.
 
In the future, applications for recognizing objects in various types of mobile devices such as smartphones and small robots are driven by batteries are becoming more and more critical. Deep learning based object recognition technology, which reduces power consumption and increases accuracy, have.
 
Meanwhile, Professor Soonhoi Ha is a representative researcher of hardware-software integrated design methodology. Currently, he is a member of the IEEE scholarship program and director of the SNU Center for Teaching and Learning and the head of the ESRC Center at Seoul National University.