Portrait
Hyojun Go
ELLIS PhD Student
ETH Zürich × Google Zürich
About Me

I am an ELLIS PhD student at ETH Zürich and Google Zürich, advised by Prof. Konrad Schindler and Dr. Federico Tombari, with co-supervision from Dr. Prune Truong and Dr. Goutam Bhat.

My research spans diffusion and flow models, few-step generation, reinforcement learning, 3D/4D generation, and world models. I study generative models at a fundamental level while using them to build 3D and 4D worlds grounded in geometry and physical structure. My recent work explores diffusion post-training with reinforcement learning and value models, as well as generative models for creating 3D and 4D worlds.

Previously, I worked as a Research Scientist at Twelve Labs and Riiid while completing my mandatory military service in Korea. Prior to that, I received my M.S. degree from KAIST.

Education
  • ETH Zürich
    ETH Zürich
    ELLIS PhD Student (with Google Zürich)
    Apr. 2025 - Present
  • KAIST
    KAIST
    M.S. in Electrical Engineering
    Mar. 2020 - Feb. 2022
  • Hanyang University
    Hanyang University
    B.S. in Electrical Engineering
    Mar. 2015 - Aug. 2019
Experience
  • Google Zürich
    Google Zürich
    Student Researcher — 3D/4D Generation and World Models
    Jul. 2026 - Present
  • Twelve Labs
    Twelve Labs
    Research Scientist (Mandatory military service) — Video Language Models (Pegasus-v1)
    Sep. 2023 - Mar. 2025
  • Riiid
    Riiid
    Research Scientist (Mandatory military service) — Diffusion Models
    Mar. 2022 - Sep. 2023
Honors & Awards
  • ICLR 2026 Oral Presentation — Top 1.6% of reviewed submissions
    2026
  • Qualcomm Innovation Fellowship 2024 Korea — Finalist (HarmonyView)
    2024
Invited Talks
  • VIST3A: Text-to-3D by Stitching a Multi-view Reconstruction Network to a Video Generator — Google Zürich
    Oct. 2025
  • Addressing Negative Transfer in Diffusion Models — AI Seoul, Seoul City Hall
    Feb. 2024
News
2026
Started as a Student Researcher at Google Zürich, working on 3D/4D generation and world models.
Jul 01
Two papers (one main, one Findings) accepted to CVPR 2026.
Feb 26
VIST3A accepted to ICLR 2026 as an Oral presentation.
Jan 22
2025
Two papers on 3D generation accepted to ICCV 2025.
Jun 15
One paper on 3DGS generation accepted to CVPR 2025.
Feb 15
One paper on Diffusion Models accepted to ICLR 2025.
Jan 15
2024
One paper on Diffusion Models accepted to AAAI 2025.
Dec 15
One paper on Diffusion Models accepted to ECCV 2024.
Jul 15

Publications

Search & filter →
Thumbnail for VIST3A: Text-to-3D by Stitching a Multi-view Reconstruction Network to a Video Generator

VIST3A: Text-to-3D by Stitching a Multi-view Reconstruction Network to a Video Generator

Hyojun Go, Dominik Narnhofer, Goutam Bhat, Prune Truong, Federico Tombari, Konrad Schindler

International Conference on Learning Representations (ICLR) 2026 Oral

Unified a pretrained video diffusion model and a feed-forward 3D reconstruction model into a single end-to-end latent diffusion model that directly generates 3D worlds from text.

PaperCodeProject

Thumbnail for Understanding, Accelerating, and Improving MeanFlow Training

Understanding, Accelerating, and Improving MeanFlow Training

Jin-Young Kim*, Hyojun Go*, Lea Bogensperger, Julius Erbach, Nikolai Kalischek, Federico Tombari, Konrad Schindler, Dominik Narnhofer

IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2026

Analyzed MeanFlow's training dynamics and developed a training scheme that improves one-step generation with 2.5× faster convergence.

Paper

Thumbnail for SteerX: Creating Any Camera-Free 3D and 4D Scenes with Geometric Steering

SteerX: Creating Any Camera-Free 3D and 4D Scenes with Geometric Steering

Byeongjun Park*, Hyojun Go*, Hyelin Nam, Byung-Hoon Kim, Hyungjin Chung, Changick Kim

International Conference on Computer Vision (ICCV) 2025

Introduced zero-shot inference-time geometric steering with reconstruction-based rewards and particle filtering for camera-free 3D/4D generation.

Paper

Thumbnail for SplatFlow: Multi-View Rectified Flow Model for 3D Gaussian Splatting Synthesis

SplatFlow: Multi-View Rectified Flow Model for 3D Gaussian Splatting Synthesis

Hyojun Go*, Byeongjun Park*, Jiho Jang, Jin-Young Kim, Soonwoo Kwon, Changick Kim

IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2025

Built a unified rectified-flow model that generates and edits 3D Gaussian Splats by jointly modeling multi-view images, depths, and camera poses.

PaperCodeProject

Thumbnail for Addressing Negative Transfer in Diffusion Models

Addressing Negative Transfer in Diffusion Models

Hyojun Go*, JinYoung Kim*, Yunsung Lee*, Seunghyun Lee*, Shinhyeok Oh, Hyeongdon Moon, Seungtaek Choi

Advances in Neural Information Processing Systems (NeurIPS) 2023

Reframed diffusion pre-training as multi-task learning and showed that harmonizing training across denoising tasks improves generation quality and convergence.

PaperCodeProject