Chanhee Lee

Sungkyunkwan University, Seoul, South Korea
B.S. in Applied Artificial Intelligence

VIP (Visual Information Processing) Lab, Seoul National University, Seoul, South Korea
Undergraduate Research Intern (Advisor: Prof. Joonseok Lee)

Efficient Learning Laboratory (ELL), Sungkyunkwan University, Suwon, South Korea
Undergraduate Research Intern (Advisor: Prof. Hankook Lee)

[P6] Well Begun Is Half Solved: Future-Aware Route Optimization for Long Narrative Question Answering, Course Project

Proposed FARO, an inference-time framework that commits to a future-aware evidence route before final answer generation for long narrative question answering.

Report

[P5] Beyond Highlight Detection: Most-Replayed Driven Multimodal Analysis of Korean YouTube Videos for Highlight Editing Guidance, Course (Capstone) Project

Built a most-replayed driven multimodal analysis pipeline for Korean YouTube videos and generated category-aware highlight editing guidance.

Project Page Report Code Dataset

[P4] Rethinking Diffusion-Based Augmentation: Why Single Prompt Fails, Course Project

Conducted a controlled study on single-prompt diffusion-based augmentation and its impact on generalization.

Report Code

[P3] N-TIDE: Debiasing Unimodal Vision Models via Neutral Text Inversion with CLIP, Course Project

Developed a CLIP-guided two-stage framework to distill debiased semantic representations into a unimodal vision model.

Report Code

[P2] BRNet: Bio-Receptor Networks for Object Detection with Zero-Shot Domain Adaptation, Undergraduate Research Program

Proposed and validated a biologically inspired luminance-adaptive detection framework for zero-shot day-to-night adaptation.

Report Code

[P1] CoReaP: Collaborative Reconstruction with Assistive Priors, AI Research Competition

Proposed a frequency-guided two-path transformer architecture for structured and detail-preserving image inpainting.

Report Code

Dacon 2025 Bias-A-Thon, Top-4

Developed prompt engineering and RAG methods for generating fair responses under biased scenarios.