Computer vision · KAIST AI

Mun-gyeom Kim

I study how we can reconstruct and understand the 3D world from images and videos.

I am an integrated M.S./Ph.D. student at CVLAB, KAIST AI. My research focuses on 3D vision and 3D/4D reconstruction, with work on compact Gaussian representations, multi-view point tracking, and spatial understanding.

Google Scholar ↗ GitHub ↗ Email ↗ CV ↗

mungyeom011@kaist.ac.kr

News

Sep 2026
Imagine3D-LLM was accepted to NeurIPS 2026.
May 2026
New preprint: T4GS, temporally aware compact Gaussians for feed-forward 4D reconstruction.
2026
C3G and MV-TAP were accepted to CVPR 2026.
Sep 2025
I joined KAIST CVLAB as an integrated M.S./Ph.D. student, advised by Prof. Seungryong Kim.
Sep 2025
OpenBox was accepted to NeurIPS 2025 as a Spotlight.

Publications

View on Scholar ↗

Conference papers. * Equal contribution (co-first authors).

2026

Concept illustration of multiple room views assembled into a 3D scene for spatial reasoning.

NeurIPS 2026

Imagine3D-LLM: Teaching MLLMs to Imagine 3D Scenes Before Answering

Jaewoo Jung, Hyeonseo Yu, Honggyu An, Jisang Han, Mungyeom Kim, Minkyeong Jeon, Heeseong Shin, Wonjun Moon, Federico Tombari, Daniel Barath, Marc Pollefeys, Seungryong Kim, Sunghwan Hong

Building compact 3D scene representations inside multimodal language models for spatial reasoning.

Project PDF arXiv Code

Concept illustration of a compact Gaussian scene reconstructed from two camera views.

CVPR 2026

Learning Compact 3D Representations from Feed-Forward Novel View Synthesis

Honggyu An^*, Jaewoo Jung^*, Mungyeom Kim, Chaehyun Kim, Minkyeong Jeon, Jisang Han, Kazumi Fukuda, Takuya Narihira, Hyunah Ko, Junsu Kim, Sunghwan Hong, Yuki Mitsufuji, Seungryong Kim

C3G learns compact Gaussian representations for efficient scene reconstruction and 3D feature lifting.

Project PDF arXiv Code

Concept illustration of corresponding colored point tracks across three views of a walking figure.

CVPR 2026

MV-TAP: Tracking Any Point in Multi-View Videos

Jahyeok Koo^*, Inès Hyeonsu Kim^*, Mungyeom Kim, Junghyun Park, Seohyeon Park, Jaeyeong Kim, Jung Yi, Seokju Cho, Seungryong Kim

Tracking points across multiple camera views through geometry-aware spatiotemporal reasoning.

Project PDF arXiv Code

2025

Concept illustration of cars and a pedestrian annotated with 3D bounding boxes in a point-cloud street.

NeurIPS 2025Spotlight

OpenBox: Annotate Any Bounding Boxes in 3D

In-Jae Lee, Mungyeom Kim, Kwonyoung Ryu, Pierre Musacchio, Jaesik Park

Automatic 3D bounding box annotation using 2D vision foundation models and object-aware refinement.

Project PDF arXiv Code

Preprints

Manuscripts and ongoing research. * Equal contribution (co-first authors).

2026

Concept illustration of three successive Gaussian poses moving a copper ball along a single connected trajectory.

arXiv 2026

Learning Temporally Aware Compact Gaussians for Feed-Forward 4D Reconstruction

Mungyeom Kim^*, Minkyeong Jeon^*, Honggyu An^*, Jaewoo Jung, Hyunah Ko, Jisang Han, Hyeonseo Yu, Donghwan Shin, Sunghwan Hong, Takuya Narihira, Kazumi Fukuda, Yuki Mitsufuji, Seungryong Kim

T4GS reconstructs dynamic scenes with temporally aware compact Gaussians, supporting camera-controllable video generation, point tracking, and dynamic scene understanding.

Project PDF arXiv Code