Our group conducts research on photorealistic rendering and its applications.


Deep Learning For Rendering

Over the past few years, there has been a significant advance in deep learning techniques and the methods have demonstrated a lot of potentials to incorporate them into other engineering domains. The deep learning techniques can be efficiently integrated into the existing rendering algorithms to enable the rendered image to become much higher quality. We aim to find specific problems in the rendering algorithms to be addressed by deep learning methods and solve the problems.

Sampling and Reconstruction for Rendering

To achieve a photorealistic image, a complex light transport equation (LTE) can be solved based on Monte Carlo (MC) ray tracing algorithm. However, it requires a lot of cost (i.e. time) to generate a noise free image. Various variance reduction techniques have been studied such as 'Importance Sampling' and 'Control Variates' to solve this problem. In addition, we can increase the rendering quality with low sample budgets using appropriate reconstruction filters.

Photorealistic Rendering in AR & VR

In spite of the technological progress of AR, immersive experience has not been achieved due to a lack of realistic appearance of virtual contents. To bring such realism into the AR, our group aims to render a virtual object indistinguishable from the real objects. We mainly study how to capture illumination to lit the virtual object and how to compute light transport between the real and virtual objects.

Acceleration for Monte-Carlo Ray Tracing