I am Hyeongju Lee (CV), a data & backend engineer with 29 months of industry experience, now researching 3D vision as a master's student at Kyushu Institute of Technology.
I build data pipelines end to end, from collection to serving, and turn research methods into working code.
Graduating with a master's in March 2027.
Project write-ups are on zoocasso.github.io; the code is here.
Data & backend · from sources to operations
The systems I built follow this path; on my site each one is drawn stage by stage, with my own part marked.
Research · a lens array lab in the browser (RIM, ICMV 2026 Best Poster Award)
A scene is captured through a lens array and rebuilt at any depth; the lab on my site runs in the browser.
Research · real-time detection and depth on lens-array images (master's thesis, in preparation)
Detect once on a real Lytro frame, then refocus only inside each box on the GPU: 16.4 ms per frame (61 fps, RTX 3090). Scene: Stanford Lytro Light Field Archive, occlusions_25 (Raj et al., 2016).
Research · image enhancement for low-light and hazy scenes (patent application in preparation)
Photos taken in the dark or in haze are made clear one image at a time; below, a dark photo from a public dataset. Data: SI-HDR (Hanji et al., SIGGRAPH 2022, CC BY 4.0).
- Data & backend, Vision Inside (2022–2024): collection, storage and serving for a KISTI-commissioned R&D platform designed by KISTI researchers, with models 3–5, which I built and put into production; and the pipeline, backend and frontend of a genomic report system (pipeline algorithm designed with the team lead).
- Research, Kyushu Institute of Technology (2025–): training-free light-field / integral-imaging reconstruction with 4 first-author conference papers (ICMV 2026 Best Poster Award), and an image-data pipeline for joint research with Hitachi, deployed on site in the Netherlands.
- Master's thesis (in preparation): real-time object detection and depth on lens-array images, 16.4 ms per frame on the GPU.
- Patent application in preparation: image enhancement for low-light and hazy scenes.
- Side project: owned the server, database and security of a community rhythm game, ~2,000 peak concurrent players, on AWS behind Cloudflare.
- Scholarship: Rotary Yoneyama Memorial Foundation Scholar (2026), selected through a Japanese-language interview.
| Project | What it is | Stack | Write-up |
|---|---|---|---|
| KISTI data platform (company code, not public) | Collection, loading and serving for a KISTI public R&D platform designed by KISTI researchers: data from a distributed crawling farm of laptops, model batches and a FastAPI admin interface; DB schema, backend integration and frontend wiring for the Java Spring prototype web app (with the in-house Java team) | Python, Selenium, MySQL, FastAPI, Java Spring, KeyBERT | details |
| Image-data pipeline, joint research with Hitachi (lab code, not public) | Written from scratch: machine-vision camera control (Spinnaker SDK / PySpin), capture, dataset preprocessing and a PyQt GUI for non-developer researchers; deployed on site in the Netherlands (Sep–Oct 2026) | Python, PySpin, PyQt | details |
| bio_fastapi | Genomic variant annotation & risk reports: genotypes joined to public references in four SQL stages; JSON API and HTML viewer, tested on SQLite and MySQL 8.0/8.4 (Vision Inside, 2022–2023; pipeline algorithm designed with the team lead) | FastAPI, SQLAlchemy, MySQL, Docker | details |
| origin-point-optimization-vcr | Code for my ITC-CSCC 2025/2026 papers: origin point found from camera parameters, 81 reconstructions → 1 (paper: 0.31 s vs an estimated 25.45 s) | Python, NumPy, CuPy | details |
| depth-estimation-using-vcr-algorithm-matlab | 2022 course project where the research started: object depth from integral images, VCR at candidate depths + FFT correlation filters, built on the course's lecture scripts; tested in Octave | MATLAB, Octave | – |
| realtime-vision-xr-device | Fog removal and low-light recovery on an XR headset, 2024 XR Device Content Makerthon (team 3DOS: owned all development and the stage demo; slides) | C++, OpenCV, Unity | – |
| wjmax-game-server | Play-log ingestion API of a community rhythm game (2023–2024, ~2,000 peak concurrent players), owning the server, database and security; the code here adds transactions, input validation, tests and CI | FastAPI, SQLAlchemy, MySQL, Docker, AWS, Cloudflare | details |
Write-ups on the site: RIM and iRIM · Origin point optimization
| Status | Venue | Paper Title |
|---|---|---|
| In preparation | Master's thesis | Real-time object detection and depth estimation on lens-array images (working title) |
| Best Poster Award | ICMV 2026 | Angular-to-Spatial Information Redistribution via Ray-based Inverse Mapping |
| Poster | ICCAS 2026 | Virtual Lens Array Synthesis from Multi-View Images via iRIM |
| Presented | ITC-CSCC 2026 | Accelerated Origin Point Optimization for Volumetric Computational Reconstruction |
| Presented | ITC-CSCC 2025 | Image Enhancement of Volumetric Computational Reconstruction Using Origin Point Optimization |
The data & backend tools are in the picture at the top. For research:



