이성길 연구실

Despite the strides made in graphics algorithms and hardware, real-time rendering of natural phenomena remains challenging. We believe there are always creative possibilities to improve images while maintaining interactive real-time performance using GPUs.

Deep Learning for rendering opens new possibilities. We understand DL as a compact nonlinear modeler of implicit representations for rendering problems. View synthesis is tackled with visibility-driven effective scene rasterization and implicit neural representations.

Rendering for DL focuses on generation of labeled images for network input. Differentiable rendering deals with inverse rendering problems to find or refine input attributes including geometry, materials, textures, and illumination using rasterization approach.

Rendering handles gigantic amounts of data. We develop practical GPU imaging pipelines where DL solves key problems and the remainder is accelerated by GPU rendering/computing. Applications include DL-based depth estimation for 3D image generation and immersive imaging.

We investigate technological aspects of Metaverse including modeling of virtual avatar, virtual environments, display algorithms, visual fatigue, and VR hardware. We improve optical accuracy of VR display and design novel pipelines with resolution-independent G-buffers.