Hyunjoon Lee

I am an integrated MS/PhD student at VGI Lab (Visual and Geometric Intelligence Lab) at Seoul National University, where I am advised by Prof. Jaesik Park.

My research interests lie at the intersection of 3D vision and robotics. I am particularly interested in how we can leverage rich scene information to enhance robot manipulation capabilities in complex environments.

Email CV Google Scholar LinkedIn Github

Education

Seoul National University (SNU)

Seoul, Korea • Integrated M.S./Ph.D. in AI

Advisor: Jaesik Park
Mar. 2024 - Present
Sungkyunkwan University (SKKU)

Suwon, Korea • B.S. in Mechanical Engineering

Summa Cum Laude (1/131), GPA: 4.39/4.5
Mar. 2018 - Feb. 2024

Publications

* equal contribution

Relation-Centric Open-Vocabulary
3D Gaussian Segmentation

Eunsung Cha*, Hyunjoon Lee*, Jaesik Park

Under Review, 2026

We propose a novel framework that reframes Gaussian segmentation as modeling pairwise relations between Gaussians, unifying instance-level coherence and query flexibility without per-scene optimization.

Project Page / arXiv / Code

👀 See the robot demo!

Occluded 3D Object Reconstruction
via Masked Multi-view Volumetric Transformer

Jinmo Kim, Namtae Kim, Hyunjoon Lee, Seungha Kim, Jaesik Park

Under Review, 2026

We propose a feed-forward multi-view 3D reconstruction framework that explicitly reasons about occlusions by reconstructing missing latent tokens across views.

OpenReview

CF3: Compact and Fast 3D Feature Fields

Hyunjoon Lee, Joonkyu Min, Jaesik Park

ICCV (Main & Demonstrations Track), 2025

We propose a method to build compact and fast 3D Gaussian feature fields by effectively compressing and sparsifying Gaussians, achieving competitive performance with significantly fewer gaussians.

Project Page / arXiv / Code

👀 See the Real World Demo!

⚡ Runs on a local laptop.

Domestic

Language-Driven Robotic Manipulation via Open-Vocabulary 3D Gaussian Scene Understanding

Hyunjoon Lee, Eunsung Cha, Jaesik Park

KCC (Outstanding Presentation Award), 2026

Aligning 2D Mask Space Consistency Using 3D Gaussian Splatting

Eunsung Cha, Hyunjoon Lee, Jaesik Park

KCC (Outstanding Presentation Award), 2026

Efficient Feature Lifting and Compression Using Pre-trained 3D Gaussians

Hyunjoon Lee, Joonkyu Min, Jaesik Park

KCC (Outstanding Presentation Award), 2025

Projects

Real2Sim VLA

A VLA model was fine-tuned on real-world scene data, and the corresponding scene was reconstructed with 2DGS and converted into a USD-compatible mesh for integration with Isaac Sim. This enables the VLA to run in a photorealistic Real2Sim environment for post-training and diverse scene-level evaluation.

Experience

Research Intern @ HuGe Lab, KAIST AI

Advisor: Beomjoon Kim

Developed a model to predict 6-DoF grasp poses using a depth sensor.
Jun. 2023 - Sep. 2023
Research Intern @ CSI Lab, SKKU

Advisor: Yusung Kim

Developed an RL agent using the PPO algorithm and transformer architecture.
Jan. 2023 - Jun. 2023
Research Intern @ CAMAS Lab, SKKU

Advisor: Jachoon Koo

Developed a mathematical model to elucidate the stiffness mechanism of the soft gripper.
Dec. 2019 - Mar. 2020

Awards & Honors

Outstanding Paper Presentation Award × 2, Korea Computer Congress (Aug. 2026)
Outstanding Paper Presentation Award, IPIU (Feb. 2026)
Honorable Mention, Research Paper Competition, Seoul National University (Dec. 2025)
Outstanding Paper Presentation Award, Korea Computer Congress (Aug. 2025)
Invited Talk, SNU Graduate of AI Research Exchange (Jun. 2025)
Top Excellence Award, Creative Research Project, Seoul National University (Jun. 2025)
Outstanding Award, Demo Competition, Seoul National University (Jun. 2025)
Invited Talk, NaverLabs (Mar. 2025)
Dean's List Award, Sungkyunkwan University (May 2021, Oct 2020)
Scholarship for Academic Excellence, Sungkyunkwan University (Sep. 2018, Mar. 2019, Mar. 2020, Mar. 2021)

Patents

CF3: Compact and Fast 3D Feature Fields, KR Patent 10-2025-0093174