Production R&D
VARCO 3D
Led the production 3D generation stack—from 1B+ native models and CUDA inference to mesh processing, UV unwrapping, and texture delivery.
Hwan Heo
Lead 3D AI Research Engineer
Building production-scale 3D generative systems from native representations and large-model training to CUDA inference and graphics pipelines.
Selected Projects
Production 3D AI systems · 4 papers · 6 invited talks
- Selected
- All
Production R&D
VARCO 3D
Led the production 3D generation stack—from 1B+ native models and CUDA inference to mesh processing, UV unwrapping, and texture delivery.
Research
CaPa
Designed Spatially Decoupled Attention to solve the Janus problem, enabling the first 4K-texture image-to-3D system and leading to VARCO 3D.
Research
Instant Pose for Hash-Grid NeRFs
Identified unstable hash-grid coordinate gradients and designed a smooth backward interpolation path for fast camera-pose refinement, published at ICML 2023.
Open-source Tool
2D GS Viewer
Built a browser viewer to inspect, edit, train, and export 2D Gaussian Splatting scenes.
Technical Study
Deep SfM for Radiance Fields
Compared VGGSfM and MASt3R for camera recovery in Gaussian Splatting and radiance fields.
Applied R&D
Neural Rendering in Game Engine
Built a 360° capture-to-engine pipeline, optimizing neural rendering for Unity and Unreal.
Research Selected Papers
-
CaPa: Carve-n-Paint Synthesis for Efficient 4K Textured Mesh Generation
Technical Report, Featured on Hugging Face Daily Papers, TLDR AI
-
DreamSurfels: Generative Gaussian Surfels for Geometrically Accurate 3D Content Creation
The 15th International Conference on ICT Convergence (ICTC), Oct 2024
Oral presentation, domestic conference -
Robust Camera Pose Refinement for Multi-Resolution Hash Encoding
The 40th International Conference on Machine Learning (ICML), Jul 2023
-
Semantic-aware Occlusion Filtering Neural Radiance Fields in the Wild
preprint, arXiv 2023
Speaking Invited Talks
-
Introduction to 3D Generative AI
Generative AI Seminar (CS.93000B) - KAIST
-
Introduction to 3D Generative AI
Machine Learning for Computer Vision - KAIST
-
Spatial 3D AI Technology and its Applications
K-GEO Festa - Ministry of Land, Infrastructure and Transport in Korea (국토교통부)
-
Radiance Fields and Spatial 3D AI for Digital Twin
National Geographic Information Institute in Korea (국토지리정보원)
-
Rasterization vs. Ray Tracing with 3D Gaussian Primitives
NCSOFT
-
Joint Optimization of Camera Pose and Accelerated Radiance Fields
NAVER Cloud
Technical Blog
Essays and technical notes on building 3D generative systems:
from representations and model training to CUDA kernels, geometry processing, and production engineering.