Who We Are

The Deep Imaging and Graphics (DIG) Lab at Kwangwoon University pursues research at the intersection of computer graphics, computer vision, and generative AI. We believe the most exciting problems sit between these fields — where physically-grounded rendering meets learned generative priors, and where simulation, reconstruction, and perception become parts of a single imaging stack.

What We Aim For

  • End-to-end imaging stack. From low-level GPU rendering pipelines to high-level learned perception and generation, we build technology along the entire imaging stack — and care that every layer is implementable, measurable, and robust.
  • Bridging classical and learned models. We treat physically-based rendering, neural representations (3DGS, NeRF), and generative diffusion models as complementary tools, and develop principled combinations such as inverse graphics with DiT priors, or medium-aware 3DGS with differentiable rendering.
  • Real systems, real impact. Our research is grounded in real applications — dental imaging, hairstyle editing, multi-camera digital twins, robotic sim-to-real — and feeds back into a shared in-house graphics engine that the whole lab uses.

Where We Came From, Where We’re Going

Over the past decade, the lab has built and shipped a custom rendering engine used in commercial medical visualization solutions, developed StyleGAN-based hairstyle editing systems, contributed to AR navigation for surgical workflows, and explored NeRF/instant-NGP based scene representations for visualization.

Our current research carries that same DNA — deep engineering combined with learning — into:

  • Modern in-house graphics engines (DX12, VXGI, advanced shadows)
  • Dental realistic inverse graphics (DiT + 3D Gaussian Splatting)
  • DiT-based controllable hair editing
  • 3D Gaussian Splatting — medium-aware models and high-speed on-the-fly reconstruction
  • Real2Sim / Sim2Real with the Genesis physics engine — ST2ST mappers, residual RL, large-scale multi-agent simulation
  • Domain-optimized segmentation via SAM 3-class attention tuning

See the Research Projects page for details on each line, including the students and concrete results.

Looking for Students Ready to Push the Boundary

We are actively looking for motivated undergraduate, master’s, and Ph.D. students who want to do hands-on research that combines deep engineering with modern AI:

  • You like building real systems — engines, simulators, training pipelines — not only writing papers about them.
  • You want to work across graphics, vision, and generative AI, instead of staying inside a single sub-area.
  • You enjoy reading the literature and getting your hands dirty in CUDA, Vulkan, PyTorch, and Genesis.

If that sounds like you, please reach out — we have open projects across all six research lines above, and we welcome new students who are ready for a challenge.

Contact: korfriend@gmail.com  |  More about the lab head