Ongoing
Probabilistic World Modeling & Text-Controllable Physics
- WorldRoPE: world-centric attention across views
- AgentDreamer: text-controllable 3D motion and physics
Hello, I'm Mijeong Kim
Ph.D. Candidate · Computer Vision Lab, SNU
My research focuses on 3D/4D reconstruction and world modeling. I develop probabilistic representations to reconstruct dynamic scenes from limited observations, with recent work extending to multi-view attention and text-controllable physics simulation.
Research Overview
2025–26
Physics-Aware Vision & Probabilistic 4D Reconstruction
- PhysGaia: physics-aware multi-body benchmark
- GP-4DGS: probabilistic 4D reconstruction from monocular video
- HyperPose: implicit pose embeddings for 3D generation
2024
Uncertainty-Aware 4D Gaussian Splatting
- UA-4DGS: uncertainty-aware 4D reconstruction from monocular video
2023
3D Scene Generation with Neural Fields
- 3D scene generation through mixtures of neural fields
2022
Few-Shot NeRF & Image Alignment
- InfoNeRF: few-shot rendering via ray-entropy minimization
- Nonparametric image alignment via graph convolutions
News
- Two first-author papers (new submissions) are under review: WorldRoPE & AgentDreamer.
- Selected as a Rising Star at the Deans’ Forum Women in Engineering Workshop.
- Two first-author papers accepted to CVPR 2026.
Publications
WorldRoPE: Probabilistic World-Centric Rotary Embedding for Multi-View Attention
Under Review
AgentDreamer: Agentic 3D Motion Reasoning for Text-Controllable Physics Simulation
Under Review
PhysGaia: A Physics-Aware Benchmark with Multi-Body Interactions for Dynamic Novel View Synthesis
CVPR 2026
GP-4DGS: Probabilistic 4D Gaussian Splatting from Monocular Video via Variational Gaussian Processes
CVPR 2026
HyperPose: Hyper-pose Embeddings for 3D-Aware Generative Models with Self-Supervised Disentangling of Pose and Scene
WACV 2026
UA-4DGS: 4D Gaussian Splatting in the Wild with Uncertainty-Aware Regularization
NeurIPS 2024
Generative Neural Fields by Mixtures of Neural Implicit Functions
NeurIPS 2023
ContraNeRF: 3D-Aware Generative Model via Contrastive Learning with Unsupervised Implicit Pose Embedding
arXiv:2304.14005
InfoNeRF: Ray Entropy Minimization for Few-Shot Neural Volume Rendering
CVPR 2022 (citation 300+)
Beyond homography: nonparametric image alignment via graph convolutional networks
MVA (Machine Vision Applications) 2022
Overcoming Forgetting in Federated Learning via Importance From Agent
Conducted in Machine Intelligence and Data Science Lab.
(430.658 of SNU, Prof. Taesup Moon)
Experience
Huawei · Noah’s Ark Labs
Sep 2025 – Feb 2026
Applied Scientist Intern · London, UK
Feed-forward 4D reconstruction from monocular video.
Kakao Brain
Jul 2022 – Sep 2022
Research Intern · Seongnam, Korea
3D-aware generation via contrastive learning · ContraNeRF
Education
-
Ph.D. in progress · Electrical and Computer Engineering
Computer Vision Lab, Seoul National University, Korea
Advisor: Prof. Bohyung Han
-
B.S. · Naval Architecture and Ocean Engineering
Seoul National University, Korea
Honors & Awards
- Youlchon AI Star – Rising Star PhD (2026)
Selected as a Rising Star PhD in artificial intelligence. - Deans’ Forum Rising Stars Women in Engineering Workshop (2026)
- Kwanjeong Ph.D. Fellowship (2019-2021)
- Science and Engineering Scholarship by President (2014-2018)
- Talent Medal of Korea, Presidential Medal (2014)
Teaching Experience
- Samsung Electronics, Korea (Jul 2019 & Jul 2020)
- Hyundai Motors, Korea (Mar 2020 - Jun 2020)
Digital Computer Concept and Practice Introduction to Python, Dept. of EE in SNU
Research Grants
Deep Learning-Based Pedestrian Removal with Multi-View Data- Fund by Naver (Sep 2019 - Aug 2020)