Jeonghwan Kim

AI Researcher / Ph.D Student @ NTU

I'm a Ph.D. student at MMLab, Nanyang Technological University (NTU), advised by Prof. Xingang Pan. I completed my M.S. in Artificial Intelligence at Konkuk University, where I was advised by Prof. Wonjun Kim. Throughout my research journey, I have primarily focused on 3D Vision and related areas in computer vision.

News

Reviewer Services:
Conference: ECCV 2026/3DV 2026/SIGGRAPH Asia 2025

Journal: Signal, Image and Video Processing • ACM Transactions on Multimedia Computing, Communications, and Applications • The Journal of Supercomputing • IEEE Transactions on Information Forensics and Security
One paper has accepted in 2026 3DV
One paper has accepted in Multimedia System
One paper has accepted in 2023 CVPR
One paper has accepted in J. Vis. Commun. Image Represent.

Publication

2025

FastMesh: Efficient Artistic Mesh Generation via Component Decoupling

J. Kim, Y. Lan, A. Fortes, Y. Chen, X. Pan.

International Conference on 3D Vision, 2026.

TL;DR: An efficient mesh generation framework that handles vertices and faces separately. Vertices are generated autoregressively, while a bidirectional transformer predicts mesh faces from vertex relationships. This reduces token redundancy to ~23% of prior methods and achieves over 8× faster mesh generation with improved quality.

Project Page arXiv Code

2023

Sampling is Matter: Point-guided 3D Human Mesh Reconstruction

J. Kim*, M. Gwon*, H. Park, H. Kwon, G. Um, W. Kim.

IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023

TL;DR: A method for 3D human mesh reconstruction from a single RGB image that samples image features at projected vertex locations to better link 2D features with 3D representation points. Combined with progressive attention masking to model local vertex interactions under occlusion, it improves reconstruction accuracy on benchmark datasets.

Paper arXiv Code

2024

Learning Scale-aware Relationships via Laplacian Decomposition-based Transformer for 3D Human Pose Estimation

J. Kim, H. Kwon, S. Y. Lim, W. Kim.

Multimedia Systems, vol. 30, Jan., 2024

TL;DR: A parameter-free 3D human pose estimation method that integrates a Laplacian pyramid with a transformer to capture multi-scale body-part relationships. It also applies body-part–wise self-attention to model local interactions, improving pose estimation performance on benchmark datasets.

Paper Code

2023

Part-attentive Kinematic Chain-based Regressor for 3D Human Modeling

J. Kim, G. Um, J. Seo, W. Kim.

Journal of Visual Communication and Image Representation, vol. 95, Sep., 2023

TL;DR: A parametric human modeling method from a single image that applies body-part attention around body centers and estimates pose and shape through a kinematic chain–based decoder. This improves robustness to severe occlusions and produces more natural human shapes and poses.

Paper Code