Smart Generation Lab

02Research

Research

We do not stop at reading about models. We train them, deploy them, and bring them into the classroom.

01 · COMPUTER VISION25 papers

Computer Vision

Image and video analysis, object recognition and tracking, character recognition, medical imaging

Where the lab began. From multi-camera object tracking and crowd behaviour analysis, through CNN variants for Korean character recognition and camera cropping for autonomous driving, to EMG-based motion estimation. The same methods were applied to medical images: cervical cell nucleus extraction and dental caries detection.

NowToday the focus is on joining vision models and generative models into one pipeline. Building the eyes that understand the world is the precondition for making new images.

Related papers · 25

17 more
02 · GENERATIVE AI4 papers

Generative AI

Image and video generation and editing, diffusion models, GANs, media synthesis pipelines

We work directly on open-source generative models and fix what needs fixing. Published work includes GAN-based facial texture control with skin condition maps, day-to-night translation with CycleGAN and attention, and zero-training image editing through pixel-level control of the MMDiT architecture.

NowA ComfyUI-based generation pipeline runs on the lab's GPU server for text-to-image, video, speech, and lip-sync video experiments. Improving generative models for cultural content production is the current project.

Related papers · 4

03 · LLM · AGENTS · EDUCATION11 papers

Others

Local LLMs and agents, reinforcement learning applications, signal processing, AI education

We push language models beyond text: Korean news summarisation and topic guiding, delta tokenisation of time series for LLM-based motion prediction, and reinforcement learning that controls game difficulty, disinfectant dosing in water treatment, and smartphone battery charging. Early work on beacon signal filtering and distance estimation sits here too.

NowWe build language models and multi-agent systems that run locally, and pass the process on through courses such as Computer Vision System Design and AI lectures for university staff.

Local LLM · Agents · 6 papers

Signal Processing · Localisation · 4 papers

AI Education

Generation,
two meanings.

What models generate, and the generation that learns to build them. Both are this lab's work.