DAEJEON, Dec. 13 (Korea Bizwire) – The Electronics and Telecommunications Research Institute (ETRI) said Thursday it will unveil the key artificial intelligence (AI) technology VoVNet (Variety of View Network), which is used to analyze visual images, as open source.
VoVNet is a backbone network, a type of computer network that interconnects various pieces of technology for implementing visual artificial intelligence such as recognizing objects.
VoVNet finds the characteristics of objects in pictures, analyzes the information, and then creates models with artificial neural networks.
Furthermore, ETRI introduced ‘SC-FEGAN’ technology that allows one to edit human faces naturally and restore their original state.
The hairstyle or facial expression of a person in an image can be modified without awkward distortions and can be restored naturally to its original appearance, even if parts of the image are damaged or missing.
Both technologies feature a GAN (Generative Adversarial Network), a deep learning technique.
A GAN is a deep learning method that allows the AI to approach a real image with learning by comparing a network that produces similar images and a network that identifies real images.
Meanwhile, the technology is expected to be used to improve the quality of results in areas such as computer graphics, web design and industrial design.
ETRI will also unveil 560 types of environmental objects and 200,000 data records on object recognition, which are needed for AI deep learning.
The data provided is from recordings of CCTV cameras on electric poles, lights, and cars. It can then be used in various fields such as urban planning, safety, environment, and transportation.
These technologies and data are available at the open source community GitHub.
“We have been using a lot of backbone network technology that foreign companies are unveiling, but it has been expensive because we need a high-performance computer,” said Dr. Park Jong-yeol of ETRI.
“This technology will help start-up businesses in related fields as it can analyze visual intelligence even with low-end computers.”
Kevin Lee (email@example.com)