Naver Again Demonstrates World-Leading AI Capability at CVPR

NAVER LABS took part in CVPR 2019, a conference on computer vision and deep learning, to share its artificial intelligence technology and recruit talent. CVPR has been co-hosted since 1983 by IEEE, the world's largest technical professional organization, and CVF. Leading research institutions attend and share new research in computer vision, deep learning and other AI technologies each year. All four papers Naver presented in the regular sessions of CVPR 2019 fall within the 'vision' area of artificial intelligence. AI vision technology is used in many areas, including translation, image classification, autonomous driving and video services.

The Clova AI researchers' paper 'Character Region Awareness for Text Detection' (Baek Young-min, Lee Bado, Han Dong-yoon, Yun Sang-doo, Lee Hwal-suk) predicts character regions using the probability of a character being present and the probability that adjacent characters belong to the same word. The technology has already been applied in Naver's services. The technology in 'An Active Learning Method for Detecting Changes in Areas of Interest for Map Updating', presented by NAVER LABS researchers, was developed over about a year of joint research by NAVER LABS teams in Korea and Europe. It analyses large-scale indoor spatial data collected by autonomous robots and recognizes changed shop signs to keep map information up to date, underpinned by NAVER LABS' core technologies in robotics, computer vision and deep learning. This year the Naver Clova team and NAVER LABS also took first place in various challenges in which global companies competed on technology. In the 'local features' category of the long-term image-based localization challenge in particular, NAVER LABS Europe took first place with a high score, well ahead of technology companies such as Google and Microsoft. The challenge involves locating the current position from night-time images on the basis of daytime images of a particular landmark and their capture locations. Using new high-performance deep learning-based techniques, NAVER LABS Europe researchers produced results surpassing SIFT (an algorithm that extracts features invariant to image scale and rotation), which had been used in the field for nearly 15 years. In the 'iFood 2019' challenge, sponsored by Google and Amazon, the Clova Vision team (Hong Ki-ho, Lee Jung-gyu, Won Tae-ryun) took first place, as did the Clova Speech team (Jung Jun-sun, Lee Bong-jin) in the ActivityNet challenge sponsored by DeepMind and Google, among others. Seok Sang-ok, CEO of NAVER LABS, said: "This CVPR was a good opportunity to show once again that the computer vision and deep learning researchers of NAVER LABS and NAVER LABS Europe are world-class," adding: "These will become the component technologies that connect high-precision data from everyday spaces with Naver services to provide new experiences and information." Kim Sung-hoon, leader of Clova AI, said: "These technologies will be applied to services soon and contribute to advancing them, delivering still greater value to users." Original article: http://m.etnews.com/20190621000179#_enliple#csidx625385323b8d785876cee769dab67ca