Vision and Video Dynamics Lab
Studying how the visual world
changes over time.
Hyeongmin Lee · SeoulTech, Dept. of Electronic Engineering
Field
Computer Vision
Image · Video
One agenda · four pillars
One question, four levels of abstraction.
Video Processing
Low-level video enhancement
Topics
- Video Frame Interpolation
- Video Super-Resolution & Enhancement
- Video Stabilization
- Video Compression
- Motion Estimation & Compensation
Enhancement
Compression
All tasks reduce to precise modeling of inter-frame motion.
Video Processing
Low-level video enhancement
Real-ESRGAN · Wang et al. ICCVW 2021
Video Processing
Low-level video enhancement
Video Processing
Low-level video enhancement
Video Understanding
Temporal semantics & representation learning
Topics
- Video Foundation Models
- Video Question Answering
- Video Retrieval
- Action Recognition
- Video–Language Alignment
- Temporal Representation Learning
From pixels to meaning — what changes, and why.
Visual reasoning
Obama is playfully pressing down on the scale with his foot to make the man weighing himself appear heavier. The aides nearby are laughing at the prank.
A group of men in suits are standing in a hallway, and one of them is using a weighing scale.
Same question, two attempts. Modern VLMs can do situational reasoning — but the difference between getting the joke and missing it is exactly what we work on.
Even simple things break.
The model says 5 fingers — but the emoji shows 4 fingers + 1 thumb. Counting and strict definitions still trip up state-of-the-art VLMs.
Glasses going up, or coming down?
No model — and no human — can tell.
Direction of motion, intent, before-and-after — these only exist between frames.
Image understanding is not video understanding. We need temporal models that reason about change.
3D / 4D Vision
3D scene reconstruction & rendering
Topics
- 3D Scene Reconstruction
- Neural Radiance Fields (NeRF)
- 3D & 4D Gaussian Splatting
- Novel View Synthesis
- Dynamic Scene Reconstruction
From 2D observations to 3D — and into time, when scenes move.
3D / 4D Vision
3D scene reconstruction & rendering
3D Gaussian Splatting for Real-Time Radiance Field Rendering · Kerbl et al. SIGGRAPH 2023
3D / 4D Vision
3D scene reconstruction & rendering
World Models
Predicting — and interacting with — the visual world
Topics
- World Foundation Models
- Interactive Video Generation
- Action-Conditioned Generation
- Latent Action Models
- Physics-Aware Generation
- Foundation Models for Embodied AI
Don't just predict the world. Act in it, and watch it respond.
World Models
Generative simulation of the visual world
World Models
Generative simulation of the visual world
DreamDojo: A Generalist Robot World Model from Large-Scale Human Videos · Gao et al. NVIDIA 2026
Why one lab, four pillars
Not four topics.
One question, four scales.
Where this leads
Where you can go after this.
… and graduate school, startups,
or anywhere visual AI is used.
Why this field
Three reasons it's a great field to be in.
All you need is an idea and a laptop. Research happens wherever you are.
The field moves fast. New questions open up every week — there is always something to chase.
A paper today can be a product people use within months. Your work reaches real users — quickly.
ViViD Lab
Come talk to us.
Research, course questions, or just curious — drop by anytime.
Scan