Gyeongsu (Bob) Cho
Research internship — Summer 2027
Seeking a Summer 2027 research internship. I study 3D/4D reconstruction of articulated humans and animals, animatable Gaussian representations, and video-based motion.
I am a Ph.D. student at UNIST, advised by Prof. Kyungdon Joo, with expected graduation in February 2028. Currently a Research Intern at UT Austin (2026–Present), working with Prof. Georgios Pavlakos and Dr. Hezhen Hu.
Published evidence
My lead-author work connects single-image animal reconstruction with temporally coherent, video-based articulated motion.
- DogRecon — International Journal of Computer Vision (IJCV), 2025. Canine-prior-guided, animatable 3D Gaussian dog reconstruction from a single image. Internet-sourced image-to-3D test: CLIP-Similarity 0.8006, measuring rendered-view semantic similarity rather than geometry accuracy. [paper, Table 2] · [project]
Earlier workshop version: CV4Animals Workshop @ CVPR 2024 (Oral). [workshop paper] · [oral video] - WildAni4D — CVPR 2026 Findings. Monocular-video animal mesh reconstruction, articulated motion, and world-space trajectories; synthetic-video and temporal reconstruction pipelines. Synthetic test set: MPJPE 91.29 and Shape Consistency 0.0703 (lower is better; units are not specified in the table). [paper, Table 1] · [project]
- AniGauss — CV4Animals Workshop @ CVPR 2026. Topology-aware Gaussians for animatable animal reconstruction from a single in-the-wild image, extending beyond dogs. The workshop paper presents qualitative comparisons across animal categories. [paper, Figure 2] · [project]
Coauthored work
- Pose-Diverse Multi-View Virtual Try-on from a Single Frontal Image — WACV 2026. [paper]
- Avatar++: Fast and Pose-Controllable 3D Human Avatar Generation from a Single Image — Wild3D Workshop @ ICCV 2025. [publication record]
- SlaBins: Fisheye Depth Estimation using Slanted Bins on Road Environments — ICCV 2023, in collaboration with 42dot. [paper] · [project]
Research demos
Research contributions & tooling
- Dynamic reconstruction: temporal modeling and evaluation that distinguish shape consistency from pose accuracy.
- Human and animal avatars: body-prior integration, controllable reconstruction, topology-aware Gaussians, and differentiable rendering.
- Synthetic supervision: rendering and synthetic-video pipelines with labeled geometry, pose, and camera trajectories.
Python · PyTorch · PyTorch3D · SMAL / parametric body models · 3D Gaussian splatting · multi-camera 3D pose.
Ongoing work & research directions
I aim to connect explicit 3D/4D body representations with reusable motion and controllable video generation. Longer-term research questions include contact-aware motion, physical simulation, and world-model integration.
- AniGauss++ — Manuscript submitted to IJCV.
- Multi-Dancer Generation — Sony × UT Austin research collaboration; Manuscript in preparation for CVPR.
- Animal Agent — Manuscript in preparation for CVPR.
- Animal Motion — Manuscript in preparation for CVPR.
Applied research
- Samsung Electronics research collaboration (2024–2025): reference-guided video anomaly detection for industrial display inspection; contributions to dataset design, model architecture, and experiments. Reference-Guided Video Anomaly Detection for Display Inspection — Manuscript submitted to WACV; awaiting decision.
- 42dot / Hyundai Motor Group research collaboration: fisheye depth estimation through SlaBins (ICCV 2023).
- Startup project (KIC & George Washington University program, 2024): a three-camera real-time 3D human pose system with motion smoothing for gyms; 50+ U.S. professional interviews and first place in the program track.
Academic service & seminars
Reviewer: CVPR 2026; ECCV 2026; CV4Animals Workshop @ CVPR 2026; 4D Vision Workshop @ CVPR 2026. [service record]
Selected seminar slides and interactive previews cover 3D vision, generative models, and human/animal reconstruction, including Agentic Coding Workflow for Research.