Projects & evidence
From single-image animal reconstruction to video-based motion: published systems, visible outputs, and my contributions. [publication record]
DogRecon
IJCV 2025; CV4Animals @ CVPR 2024 oral [citation]
What I built: Single-image dog reconstruction with canine-prior-guided novel views and an animatable Gaussian representation. [project]
My contribution: Lead author; developed the reconstruction framework and canine-prior / reliable-sampling research direction.
What it showed: Internet-sourced image-to-3D test: CLIP-Similarity 0.8006. This is a rendered-view semantic-similarity metric, not geometry accuracy. [paper, Table 2]
WildAni4D
CVPR 2026 Findings [citation]
What I built: Video-native recovery of animal mesh, articulated motion and world-space trajectory. [project]
My contribution: Lead author; developed synthetic-video and temporal reconstruction research pipelines.
What it showed: Synthetic test set: MPJPE 91.29 and Shape Consistency 0.0703 (lower is better). These are paper-reported values; units are not specified in the table. [paper, Table 1]
AniGauss
CV4Animals Workshop at CVPR 2026 [citation]
What I built: Topology-aware Gaussians for animatable animal reconstruction from a single in-the-wild image. [project]
My contribution: Lead author; developed the animal reconstruction research line extending beyond the dog-specific setting.
What it showed: The published workshop manuscript reports qualitative comparisons across animal categories; no quantitative performance claim is made here. [paper, Figure 2]
Industry & applied research
Samsung Electronics — reference-guided video anomaly detection
Research collaboration, 2024–2025. Co-developed a framework for industrial display inspection with real manufacturing data; contributed to dataset design, model architecture, and experiments.
Manuscript submitted to WACV; awaiting decision.
42dot / Hyundai Motor Group — fisheye depth
Research collaboration, 2022–2023. Contributed to SlaBins, connecting fisheye camera geometry with depth estimation for road environments; the associated depth-map patent is an application, not a granted patent. [paper / code]
Sony × UT Austin — Multi-Dancer Generation
Research collaboration, not a Sony employment claim. Ongoing work on multi-dancer generation; Manuscript in preparation for CVPR.
Startup project — real-time 3D human pose
Built a three-camera system with motion smoothing for real-time pose analysis in gyms. Conducted 50+ interviews with fitness and healthcare professionals in the U.S. to refine the product direction; ranked first in the KIC & George Washington University program track.
Skills & tooling
Python · PyTorch · PyTorch3D · SMAL / parametric body models · 3D Gaussian splatting · synthetic-video and rendering pipelines · multi-camera 3D pose. [systems]
Agentic coding workflow for research: [seminar slides].
Recognition, teaching & service
AI Tech Open Workshop, UNIST: startup track first place and KRW 30,000,000 research funding; two patent applications and teaching / mentoring experience.