In this work, we propose modeling, algorithms, and simulation methods, which allow us to effectively reflect the imperfection of the camera-based real-world imagery and to utilize them for training deep-learning models. Precisely, we model the imperfection of real-world optics so that the optical quality of the virtual imagery is significantly similar in quality to real photos. We also analyze how the optical imperfections affect deep learning in terms of prediction accuracy and learning efficiency, and then, prioritize them for effective learning. Lastly, we identify the feasibility of our approach in potential applications including autonomous vehicles, aerial-view imaging, and microscopic imaging.

This research aims at developing a new GPU pipeline to address problems in costly very-high-resolution rendering. In this work, we attempt to minimize the resolution dependency of the rendering pipeline, and to design a resolution-adaptive postprocessing framework.

본 세부과제에서 구현하고자 하는 바는 맞춤형 정신 건강 증진 콘텐츠를 위한 가상현실 시스템 이다. 적정 가상현실 기술을 이용하여 의학적 지식을 바탕으로 한 맞춤형 콘텐츠의 구현을 위한 시스템을 개발하는 것을 목표로 한다. 또한, 증상별, 단계별 맞춤형 콘텐츠의 지원을 위한 보조 기술들을 개발한다.

웹툰을 위한 다양한 이미지 component들간의 layout retargetting과 그 지원을 위한 GUI 기술을 연구한다.

본 연구과제는 초고속 전역조명 기술 개발을 목표한다. 본 연구는 가상과 현실의 시각적 괴리 현격히 줄이는 초고속 전역조명 기법을 개발한다.

모바일 VR 플랫폼 상에서의 고성능 렌더링 구조와 기법을 개발하고, 이를 바탕으로 스테레오 실사 비디오 기반의 렌더링 기법을 개발한다.

This research aims at developing methods and algorithms for seamless GPU computing, which allows users to analyze, detect, and avoid severe performance seams of GPU software, and thereby, achieves high performance. Major topics include: definition of seams in GPU computing as a factor of performance drop; studies of analysis/detection of performance seams in input software; techniques and algorithms of removing seams for high-performance GPU software.

본 연구과제는 현실과 가상이 통합된 실감교류확장공간에서 공존감 제공을 위한 초고속 전역조명 기술 개발을 목표한다. 가상과 현실의 시각적 괴리는 시각적 렌더링의 물리적 사실감 부족에 기인해왔으며, 본 연구는 이러한 괴리를 현격히 줄이는 초고속 전역조명 기법을 개발하며, 세부 목표는 다음을 포함한다.

This project first analyzes reasons and effects of degradation in VR display, and then, aims to develop technologies to reduce the quality degradation in VR display. In particular, the project focuses on the optical degradation and motion blur.

This project aims to develop a high-performance 3D graphics library for coexistent reality. The objectives of the second stage include: (1) Multi-GPU based parallel processing 3D GPU library for full-HD resolution; (2) Real-time rendering for user-avatar interaction.

With the strides of high-performance GPUs, general-purpose GPU computing has been of interests, yet its accessibility and the optimal performance have still been limited. This project aims to develop a mapping model between CPU and GPU algorithms and an intuitive user interface for writing GPU software. It is distinguished from the existing low-level programming for its high-level approach with modular low-level components.

The paradigm of the recent computing is rapidly evolving to the cloud computing. Many researches in Korea is going on, but the majority of the core software is still relying on the foreign software. Hereby, this project aims to develop key technologies for cloud computing, in particular for servers, networking, and terminals.

Digital contents are expanding beyond the classic use with the strides of internet and diverse mobile platforms. Albeit a number of budgets are invested in creating the contents, their efficiency is limited by ineffective visual design without consideration on human visual perception. Among such multimedia contents, this project particularly concerns about images, videos, and 3D graphics. We aim to develop (1) a method to quantify the effectiveness of visualization, based on neuroscience studies on visual attention, and (2) its optimization algorithms.