PeopleofSuperintelligence

Researcher

Jaesung Choe

Also known as 최재성

Jaesung Choe is a research scientist at NVIDIA Research Taiwan who works on 3D computer vision and on multimodal language models for 3D data. Before completing a PhD in 2023 in the Robotics and Computer Vision Lab of KAIST's Department of Electrical Engineering, he coauthored Segment2Regress, VolumeFusion and PointMixer with In So Kweon. At NVIDIA, his work with collaborators includes the 4DRegSDF method for dynamic scenes, sparse voxel rasterization for real-time rendering, the Mosaic3D dataset of 5.6 million 3D mask-text pairs, and Dr. Splat, which attaches language embeddings to 3D Gaussians. Papers from 2026 such as OpenVoxel and SplatReasoner help vision-language models reason over 3D scenes, and he is building a foundation model for embodied agents.

Timeline

  1. 2019

    Segment2Regress at RSS

    He publishes Segment2Regress at Robotics: Science and Systems 2019 with Kyungdon Joo, Francois Rameau, Gyu Min Shim and In So Kweon. The two-stage network locates vehicles in 3D from a single camera image and reaches state-of-the-art results on the KITTI bird's eye view dataset.

  2. 2021

    VolumeFusion for 3D scenes

    At ICCV 2021, he and coauthors including In So Kweon present VolumeFusion, a deep network that fuses depth maps and image features into a single TSDF volume to reconstruct 3D scenes. The method adds PosedConv, a rotation-invariant 3D convolution kernel, and is tested on the ScanNet dataset.

  3. 2022

    PointMixer and point cloud reconstruction

    In 2022, he publishes PointMixer, an MLP-only architecture for 3D point cloud understanding, at ECCV 2022, and Deep Point Cloud Reconstruction, which upsamples and denoises point clouds jointly, at ICLR 2022. Jaesik Park and In So Kweon coauthor both papers.

  4. 2023

    A doctorate at KAIST

    He completes a PhD in 2023 in the Robotics and Computer Vision Lab of KAIST's Department of Electrical Engineering and moves to NVIDIA.

  5. September 2023

    4DRegSDF for dynamic scenes

    In September 2023, he publishes 4DRegSDF at ICCV 2023 with Christopher Choy, Jaesik Park, In So Kweon and Anima Anandkumar. The algorithm imposes local rigidity on a deformable signed distance function to improve neural rendering and reconstruction of dynamic scenes.

  6. 2025

    Three CVPR 2025 papers

    At CVPR 2025, he coauthors Sparse Voxels Rasterization, a radiance field renderer that rasterizes adaptive sparse voxels without neural networks or 3D Gaussians. He also coauthors Mosaic3D, a dataset of 5.6 million 3D mask-text pairs from more than 30K scenes with a matching open-vocabulary 3D segmentation model. As a corresponding author of Dr. Splat, a CVPR 2025 highlight, he helps link CLIP language embeddings directly to 3D Gaussians.

  7. 2026

    OpenVoxel and SplatReasoner

    OpenVoxel, written with Sheng-Yu Huang, Yu-Chiang Frank Wang and Cheng Sun, appears at CVPR 2026; it groups and captions sparse voxels without training so that multimodal language models can search a 3D scene by text. SplatReasoner, at ECCV 2026, uses 3D Gaussian Splatting to render new viewpoints that improve the embodied reasoning and 3D grounding of vision-language models. Two more of his papers are accepted to NeurIPS 2026 in September 2026.

  8. As of October 2026

    Research scientist at NVIDIA Taiwan

    He is a research scientist at NVIDIA Research Taiwan, where his main research interest is multimodal large language models for 3D data. He is building a foundation model for embodied agents and mentors research interns, starting with POSTECH student Chunghyun Park in 2023.

  1. Google had the most people shaping AI every year from 2014 to 2025Our estimates of which organizations employed the most people shaping AI from 1961 to 2025, with a year-by-year chart.
  2. The United States employs an estimated 53% of people in our data with a known employment countryOur estimates of where AI careers are concentrated now and how countries’ shares of dated positions changed over time.
  3. 46% of living people in our data with known birth and employment countries work outside their birth countryOur estimate of cross-border careers in AI, from country of birth and the country of each current organization.
  4. More people shaping AI studied at Massachusetts Institute of Technology than at any other institutionOur estimates of where the people shaping AI studied and where their careers led, from education records in cited biographies.

Colleagues

People named in this biography, and people with dated positions at the same organization in the same years.

Sources and credits

Photo: We have no published photo of Jaesung Choe yet. Send us a free-licensed photo.

  1. Segment2Regress: Monocular 3D Vehicle Localization in Two Stages (Robotics: Science and Systems XV) roboticsproceedings.org. Checked 19 October 2026.
  2. VolumeFusion: Deep Depth Fusion for 3D Scene Reconstruction (ICCV 2021 Open Access Repository) openaccess.thecvf.com. Checked 19 October 2026.
  3. Jaesung Choe jaesung-choe.github.io. Checked 19 October 2026.
  4. Alumni | Robotics and Computer Vision Lab, KAIST rcv.kaist.ac.kr. Checked 19 October 2026.
  5. Spacetime Surface Regularization for Neural Dynamic Scene Reconstruction | NVIDIA Research Taiwan research.nvidia.com. Checked 19 October 2026.
  6. [2412.04459] Sparse Voxels Rasterization: Real-time High-fidelity Radiance Field Rendering arxiv.org. Checked 19 October 2026.
  7. Mosaic3D: Foundation Dataset and Model for Open-Vocabulary 3D Segmentation nvlabs.github.io. Checked 19 October 2026.
  8. Dr. Splat: Directly Referring 3D Gaussian Splatting via Direct Language Embedding Registration drsplat.github.io. Checked 19 October 2026.
  9. OpenVoxel peterjohnsonhuang.github.io. Checked 19 October 2026.
  10. SplatReasoner: Enhancing Embodied Reasoning and Grounding by Novel View Synthesis splatreasoner.github.io. Checked 19 October 2026.
  11. Jaesung Choe | NVIDIA Research Taiwan research.nvidia.com. Checked 19 October 2026.