Dongho Lee
M.S. Researcher · Dong-A University · ISPL

Dongho
Lee

Researching Video Super Resolution, video compression standards including HEVC, VVC, and Beyond VVC, Rate Control algorithms, and 3D Gaussian Splatting at Image Signal Processing Lab, Dong-A University. Advised by Prof. Dongsan Jun.

SCIE (Under Review) KCI Publication 3 Patents Filed Best Paper Award NIA Gov. Project
Profile

About Me

I am a M.S. student in the Department of Computer Engineering at Dong-A University, Busan, South Korea, advised by Prof. Dongsan Jun in the Image Signal Processing Laboratory (ISPL).

My primary research interest lies in deep learning–based video quality enhancement. I have focused on Video Super Resolution (VSR) using deformable convolution–based alignment and dense feature extraction, targeting both quality-maximizing and edge-device–deployable lightweight architectures.

In parallel, I work on VVC (H.266) video compression, focusing on inter-prediction optimization and rate control algorithms — optimizing QP scheduling and bit allocation under constrained bitrate budgets. This line of work has led to 3 domestic patent filings.

More recently, I have expanded into 3D Gaussian Splatting (3DGS), exploring scene analysis and quality evaluation for real-time novel-view synthesis pipelines. I also contributed production-level software to the Busan Smart City Digital Twin Platform, commissioned by the National Information Society Agency (NIA).

2
Publications
(Journal + Conference)
3
Patents Filed
(Korean Intellectual Property)
5
Open-Source
Projects
2
Awards
(Best Paper + Capstone)

Focus Areas

Research Interests

Video SR

Video Super Resolution

Video Super Resolution (VSR) is a class of methods that reconstruct high-resolution video frames from temporally consecutive low-resolution inputs by exploiting inter-frame spatial-temporal correlations through deep neural networks.

3D Rendering

3D Gaussian Splatting

3D Gaussian Splatting (3DGS) is a point-based scene representation that models a scene as a set of anisotropic 3D Gaussians, enabling real-time novel-view synthesis via differentiable rasterization without neural ray marching.

Video Coding

VVC / Video Compression

Versatile Video Coding (VVC / H.266) is the latest generation ITU-T/ISO video coding standard that achieves approximately 50% bitrate reduction over HEVC through advanced partitioning, inter/intra prediction tools, and in-loop filtering.

Rate Control

VVC Rate Control

Rate control in video coding determines the quantization parameter (QP) and bit allocation per coding unit to meet a target bitrate constraint while maximizing perceptual quality, balancing distortion against buffer and bandwidth limits.

Digital Twin

Smart City Digital Twin

A smart city digital twin is a real-time virtual replica of urban infrastructure — integrating GIS, IoT sensor streams, and 3D city models — that enables simulation, monitoring, and data-driven decision support for city administrators.

Image Quality

Perceptual Image Quality

Image quality assessment (IQA) encompasses both full-reference metrics (PSNR, SSIM, VMAF) and no-reference perceptual models that measure fidelity, structural similarity, and human visual system–aligned distortion in restored imagery.


Academic Output

Publications

SCIE Journal
TBD
International Journal · SCIE Under Review
Title to be updated upon submission
Dongho Lee, Dongsan Jun, et al.
Submitted to an SCIE-indexed international journal
VVC Rate Control
KCI Journal
2023
Journal · KCI Indexed
Video Super Resolution Method using Deformable Convolution based Alignment Network with Residual Dense Block
Dongho Lee, Yooho Lee, Sejin Chun, Dongsan Jun
Journal of Korea Multimedia Society, Vol. 26, No. 5, pp. 650–659, May 2023  ·  DOI: 10.9717/kmms.2023.26.5.650
VSR Deformable Convolution Residual Dense Block REDS Vimeo-90K +0.23 dB PSNR vs DCAN
Domestic Conference
2023
Conference · 한국방송미디어공학회 하계학술대회
Video Super Resolution Method using Deformable Convolution based Alignment Network with Lightweight Feature Extraction Block
Dongho Lee, Byungju Park, Youngwoo Lee, Sukhee Cho, Jangwu Jo, Dongsan Jun
2023 KIBME Summer Conference  ·  Supported by IITP (No. 2021-0-00087)
VSR Lightweight CNN LDCAN −38% Parameters vs DCAN Edge Device

Intellectual Property

Patents

KR Patent
10-2025-0193072 · Filed 2025.12.08
Method for Augmenting TMRL Candidate List
TMRL candidate 추가 구성 방법 · Inventors: Dongsan Jun, Dongho Lee · Dong-A Univ. Industry–Academia Cooperation Foundation
Pending
KR Patent
10-2025-0193071 · Filed 2025.12.08
Method for Augmenting a TIMD Merge List
TIMD merge list 추가 구성 방법 · Inventors: Dongsan Jun, Dongho Lee · Dong-A Univ. Industry–Academia Cooperation Foundation
Pending
KR Patent
10-2024-0071658 · Filed 2024.05.31
Template Matching-based Pair-wise Average Motion Vector Prediction Candidate List Extension
Template matching 기반 pair-wise average MVP 후보 목록 확장 방법 · Inventors: Dongsan Jun, Taesik Lee, Dongho Lee
Pending

Selected Work

Projects

Deep Learning

VideoSR — RDAN

Video Super Resolution using Residual Dense Alignment Network (RDAN). Achieves +0.23 dB PSNR improvement over DCAN on REDS4 benchmark. Published in KCI Journal of Korea Multimedia Society (2023).

Python PyTorch REDS Vimeo-90K Tesla V100
3D Vision

3DGS-Analyzer

Analysis and evaluation toolkit for 3D Gaussian Splatting scenes. Supports scene statistics, quality metrics, and visualization of Gaussian primitives for research in novel-view synthesis.

Python 3D Gaussian Splatting Novel View Synthesis
Gov. Project

Light Pollution Simulation

Front-end for Busan Metropolitan City's 1365 Smart City Digital Twin Platform ("Urban Artificial Lighting Safety Service"), commissioned by NIA. Features real-time 3D heatmap, light-source simulation, and admin dashboards. Live at 1365twin.busan.kr.

3D Digital Twin WebGL GIS Real-time
NIA Gov. Project · Busan Metropolitan City
LiDAR

Pedestrian Safety System

LiDAR-based obstacle detection system providing directional audio warnings for visually impaired pedestrians. Won Capstone Design Award at Dong-A University. Research paper presented at IEEE conference.

3D LiDAR Point Cloud Obstacle Detection Audio Guidance
Unreal Engine

Interactive Planetarium (VR)

3D interactive planetarium built in Unreal Engine — explore and learn about 52 constellations in virtual reality. Demonstrates real-time 3D rendering and interactive XR design.

Unreal Engine C++ VR 3D Interactive

Recognition

Awards & Honors

Best Paper
2024
최우수논문상 (Best Paper Award)
한국멀티미디어학회 (Korea Multimedia Society) · Dong-A University
Capstone
Capstone Design
Capstone Design Award — Pedestrian Safety System
Dong-A University · LiDAR-based Walking Assistance System for Visually Impaired
NIA
2023 – Present
NIA Government-Commissioned Project Contributor
National Information Society Agency (NIA) · Busan 1365 Digital Twin Platform

Background

Education

2023 – Present
M.S. in Computer Engineering
Dong-A University, Busan, South Korea
Advisor: Prof. Dongsan Jun  ·  Lab: Image Signal Processing Lab (ISPL)
Research: Video Super Resolution, 3D Gaussian Splatting, Video Coding
– 2023
B.S. in Computer Engineering
Dong-A University, Busan, South Korea
Capstone Design Award recipient

Technical Stack

Skills

Deep Learning / ML

PyTorch Python CNN Design Deformable Conv Attention Mechanism Dense Connections

Computer Vision / Graphics

3D Gaussian Splatting Video SR Image Quality (PSNR/SSIM) LiDAR / Point Cloud WebGL GIS Visualization

Systems & Tools

Unreal Engine C++ Digital Twin NVIDIA GPU (V100) Git / GitHub Linux