GaussExplorer: 3D Gaussian Splatting for Embodied Exploration and Reasoning
arXiv, 2026
Embodied reasoning on 3D scenes using 3D Gaussian Splatting. + Academic collaboration with KAIST and POSTECH.
Jaesung Choe
Hi, I am a research scientist at NVIDIA Research Taiwan. My main research interest is 3D scene understanding and interactions using pointclouds, 3D Gaussians, and digital humans.
Email / CV / Google Scholar / LinkedIn
News
- Nov 2025: NVIDIA Research Taiwan is currently hiring for a full-time research position.
- Feb 2025: Three papers have been accepted to CVPR 2025.
Research
I'm interested in computer vision for 3D scene understanding. Representative papers are highlighted.
MV-SAM: Multi-view Promptable Segmentation using Pointmap Guidance
arXiv, 2026
A framework for multi-view promptable segmentation that achieves 3D consistency using pointmaps—3D points reconstructed from unposed images by recent visual geometry models, such as VGGT, PI3, etc. + NVIDIA research paper.
OpenVoxel: Training-Free Grouping and Captioning Voxels for Open-Vocabulary 3D Scene Understanding
arXiv, 2026
We attach natural language descriptions to 3D sparse voxels — enabling any VLM or LLM to directly reason over 3D scenes. + NVIDIA research paper.
SVRecon: Sparse Voxel Rasterization for Surface Reconstruction
arXiv, 2025
We strengthen voxel-wise associations among parent-child and sibling voxel groups, resulting in smoother surface reconstruction. + Academic collaboration with Seoul National University.
Quantile Rendering: Efficiently Embedding High-dimensional Feature on 3D Gaussian Splatting
arXiv, 2025
An efficient rendering algorithm for 3D Gaussians that involve high-dimensional feature vectors. Our Q-render achieves ∼43.7× speed gains compared to recent studies when rendering 512-D feature maps. + NVIDIA research paper.
Dr. Splat: Directly Referring 3D Gaussian Splatting via Direct Language Embedding Registration
CVPR, 2025 (highlight Top 3.7%)
Distill language knowledge into 3D Gaussian Splatting. + Winner of Qualcomm Innovation Fellowship Korea + Academic collaboration with POSTECH.
Mosaic3D: Foundation Dataset and Model for Open-Vocabulary 3D Segmentation
CVPR, 2025
Introduce a data generation pipeline that leverages 2D foundation models to synthesize 3D mask-text paired data. + NVIDIA research paper.
Sparse Voxels Rasterization: Real-time High-fidelity Radiance Field Rendering
CVPR, 2025
Extend the rasterization technique into the sparse voxel representation for the image rendering task. + NVIDIA research paper.
Spacetime Surface Regularization for Neural Dynamic Scene Reconstruction
ICCV, 2023
Apply spacetime surface regularization technique for 4D surface reconstruction and dynamic scene rendering. + NVIDIA research paper.
[Paper]
PointMixer: MLP-Mixer for Point Cloud Understanding
ECCV, 2022
Design a new MLP-only architecture for 3D points.
Deep Point Cloud Reconstruction
ICLR, 2022
Perform the point cloud upsampling and denoising tasks simultaneously. Demonstrate good generalization performance.
[Paper]
VolumeFusion: Deep Depth Fusion for 3D Scene Reconstruction
ICCV, 2021
Propose deep learning based depth fusion algorithm for indoor scene reconstruction.
[Paper]
Volumetric Propagation Network: Stereo-LiDAR Fusion for Long-Range Depth Estimation
RA-L, 2021
Estimation dense depth maps by fusing LiDAR points and stereo images.
[Paper (RA-L)] [Paper (ICRA)] [Video]
Mentorship
I have mentored the following research interns.- 2026 Yu-Ji Kim (POSTECH)
- 2024 Yoonwoo Jeong (POSTECH)
- 2023 Chunghyun Park (POSTECH)
Academic collaboration
For research collaboration, I have collaborated with the following researchers.- 2026 Open (feel free to reach out if interested!)
- 2025 Seunghun Oh (Seoul National University)
- 2025 Yu-ji Kim (POSTECH → with me → NVIDIA intern)
- 2024 Junseong Kim (POSTECH → with me → Huawei intern)