Deep Learning & Computer Vision Engineer
Building reliable 2D/3D perception systems and generative models for real-world applications.
(Currently restructuring repositories β demos, evaluation scripts, and project pages are being added progressively.)
Computer Vision Β· 3D Generative Models Β· Generative AI Β· Probabilistic ML
Python Β· PyTorch Β· C/C++ Β· CUDA Β· FastAPI Β· Docker Β· AWS Β· Linux
(More structured READMEs, demos, and weights will be available as repositories are updated)
Feature-space repulsion (DINOv2) + 3DGS for diverse and stable text-to-3D generation.
- ~98% β semantic diversity
- Fidelity preserved (ΞCLIP β β0.006)
- Multi-view consistency C > 0.83
- Human perceptual study (n = 41)
- Scalable N-parallel pipeline with reproducible experiments
π Repo:sijeong-kim/3D-Generation(demo uploading soon)
Fusion of 2D lifting + temporal transformer models (VideoPose3D, MHFormer, MixSTE, P-STMO) for truncation-aware pose estimation.
- ~37% β MPJPE on Human3.6M under truncation
- Integrated & deployed at KIST (Koreaβs national flagship research institute)
- Some implementation details remain private due to institutional policy
π Repo:sijeong-kim/Truncation-Robust-3D-Human-Pose-Estimation(limited release)
Real-time multimodal storytelling (LLM + diffusion + text-to-music).
- LoRA-tuned Stable Diffusion & HyperCLOVA X
- FastAPI + Nginx + GPU pipeline
- Excellence Award (2023)
π Repo:6garlics/tori-ai
(Live service was deployed in 2023, now archived but code fully available.)
- MSc Computing (AI & ML), Imperial College London β Distinction
Thesis: Diversify Guided 3D Generation via Repulsive 3D Gaussian Splatting - BSc Computer Science, Ewha Womans University β Summa Cum Laude (GPA 4.2/4.3)