
Decoupled Reprojection Consistency for Diagnosing 3D Gaussian Splatting Failures
Eurographics 2026 Poster
I am a Research Engineer at Neowiz and received my M.S. from Chung-Ang University. My work centers on learning-based visual representations with an emphasis on physical structure and controllability.
I am interested in how neural models can be guided by physically meaningful priors such as geometry, materials, and intrinsic image structure to support reliable reconstruction, rendering, and editing of visual content. In practice, I care most about systems that survive real production workflows and remain useful to artists and downstream tools.

Eurographics 2026 Poster

Neowiz, Jan 2024 - Present
Developed and iteratively improved a studio-specific AI image generation workflow for concept artists, rapidly integrating new deep learning methods for IP-consistent concept art generation, style transfer, and background variation.

Neowiz, Apr 2024 - Nov 2024
Designed a hair guide model generation pipeline for game hair production using geometry processing, clustering, importance sampling, and resampling.

Neowiz, Aug 2022 - Oct 2022
Designed and implemented an automated pipeline for facial animation generation using speech-to-viseme and script-based sentiment cues.

Neowiz, Jun 2021 - Oct 2022
Conducted GAN-based domain transfer to convert monster voices into mechanical sounds; applied in Lies of P.

National Research Foundation of Korea, Sep 2018 - Oct 2019
Conducted 3D volumetric data classification using 3D convolutional neural networks, achieving high accuracy in multi-class prediction tasks.
I draw inspiration and motivation from visual things. Visual computing and painting are how I give form to my reflections.