Junsu Kim (김준수)

I am a Ph.D student in the 3D Vision & Robotics Lab, Department of Artificial Intelligence at UNIST, advised by Kyungdon Joo. Previously, I worked as an intern at 42dot Inc. and the BIG (Bot Intelligence Group) of the Robotics Institute at CMU, advised by Jean Oh. I received B.S. from the Department of Automobile and IT Convergence at Kookmin University, advised by Sanghun Lee.

My research interests lie in 3D Vision and Robotics, particularly in solving real-world problems related to robot vision.

News

Jul 02, 2025	My paper “VPOcc: Exploiting Vanishing Point for 3D Semantic Occupancy Prediction” has finally been accepted to IROS 2025!
May 22, 2025	I’ll be presenting my poster at the Foundation Models for V2X-based Cooperative Autonomous Driving workshop @ CVPR 2025. Come by and say hi!
Feb 20, 2025	I received the Silver Prize🏆 at the 31^st Samsung Humantech Paper Award ($7000)!

Publications

Vision

VPOcc: Exploiting Vanishing Point for 3D Semantic Occupancy Prediction

Junsu Kim, Junhee Lee, Ukcheol Shin, Jean Oh, and Kyungdon Joo

IROS, 2025

🏆Silver Prize at the Samsung Humantech Paper Award

Abs arXiv Website

Camera-based 3D occupancy prediction using vanishing points as geometric cues to mitigate information imbalance in perspective projection.

Projects

Vision

TL-3D: Tooth Loss Dataset and 3D Detection

Junsu Kim

On-going, 2024

Abs

3D tooth loss dataset and benchmark, and a 3D missing tooth detection model for autonomous dental implant surgery.
Vision

MonoFish: Depth-aware Monocular Fisheye 3D Object Detection

Junsu Kim

On-going, 2024

Abs

Depth-aware monocular 3D object detection model for fisheye cameras using the real-world KITTI-360 dataset to compensate for fisheye distortion.
Robot

Bimanual Robot Arm VR Teleopeariton

Junsu Kim, Tai Inui, and Uksang Yoo

CMU Research Internship, 2024

Abs

VR teleoperation system with augmented 3D reality for collecting manipulation data using dual robot arms.
Vision

Fisheye-Pinhole Camera Fusion for Multi-view 3D Object Detection

Junsu Kim

42dot Intership, 2024

Abs

Multi-view 3D object detection via fisheye-pinhole camera fusion, utilizing a pre-installed fisheye camera to enhance pinhole-based detection.
Simulation

DeepPAVE: Deep learning-based Personalized Autonomous VehiclE

Team HAVI (Project leader)

The 9th Cloud ProgrammingWorld Cup, 2021, 2021

🏆Honorable Judge Award

Abs Video Website

Personalized autonomous driving model using an LSTM-based network, trained on highway driving data collected from UC-win Road simulation and a K7 module via CAN communication.