Seonwook Park
Also known as Wookie
Seonwook Park is a computer vision researcher whose work spans gaze estimation, computational pathology, human perception, and 3D avatar systems. Originally from Seoul, he earned an MSc in Computational Science and Engineering from ETH Zurich and a BSc in Physics from Imperial College London before completing a computer science PhD there in June 2020 under Otmar Hilliges. At ETH, he coauthored the 2018 Deep Pictorial Gaze Estimation paper and the 2020 ETH-XGaze dataset, which gathered over one million high-resolution gaze images under extreme poses. He joined Lunit as a research scientist in July 2020, working on microscope image analysis and computational pathology; in 2023, he coauthored OCELOT, a CVPR dataset with overlapping cell and tissue annotations for cell detection. Since June 2024, he is a Senior Research Scientist at NVIDIA Research in the AI-Mediated Reality and Interaction group, where he works on computer vision and human perception; he coauthored GAIA in 2025 for expression-conditioned 3D head avatars and Grounded-Exo2Ego in 2026 for first-person video generation.
Timeline
-
2018
Deep Pictorial Gaze Estimation
In 2018, he coauthors Deep Pictorial Gaze Estimation. Its sequential neural network predicts a pictorial representation of 3D gaze direction before estimating gaze from a single eye image.
-
2020
PhD at ETH Zurich
Originally from Seoul, he holds an MSc in Computational Science and Engineering from ETH Zurich and a BSc in Physics from Imperial College London. He completes his computer science PhD at ETH Zurich in June 2020 under Otmar Hilliges.
-
2020
The ETH-XGaze dataset
In 2020, he coauthors ETH-XGaze, a dataset of over one million high-resolution images under extreme head poses and gaze directions. The paper receives an ECCV Spotlight presentation.
-
July 2020
Research Scientist at Lunit
He joins Lunit as a research scientist in July 2020 and works on microscope image analysis in computational pathology.
-
2023
OCELOT for cell detection
In 2023, he coauthors OCELOT, a CVPR dataset that pairs overlapping cell and tissue annotations for cell detection in pathology images.
-
2025
GAIA head avatars
In 2025, he coauthors GAIA, which generates expression-conditioned 3D head avatars for interactive animation and rendering.
-
2026
Grounded-Exo2Ego video generation
In 2026, he coauthors Grounded-Exo2Ego, a video diffusion method with two branches that generates first-person video from third-person footage using 3D reconstruction and semantic grounding.
-
June 2024–October 2026
NVIDIA Research Scientist
In June 2024, he joins NVIDIA Research as a Senior Research Scientist. He works in its AI-Mediated Reality and Interaction Research Group on computer vision, machine learning, and computational human perception.
Career
- Since 2024NVIDIASenior Research Scientist
- 2021–2024Sungkyunkwan University
- 2020–2024LunitResearch Scientist
Education
- 2020ETH ZurichMSc in Computational Science and Engineering
- Imperial College LondonBSc in Physics
Most-cited works
- ETH-XGaze: A Large Scale Dataset for Gaze Estimation Under Extreme Head Pose and Gaze VariationWith Otmar Hilliges
- Cross-Modal Deep Variational Hand Pose EstimationWith Otmar Hilliges
- Artificial Intelligence–Powered Spatial Analysis of Tumor-Infiltrating Lymphocytes as Complementary Biomarker for Immune Checkpoint Inhibition in Non–Small-Cell Lung Cancer
- Few-Shot Adaptive Gaze EstimationWith Otmar Hilliges and Shalini De Mello
- Deep Pictorial Gaze EstimationWith Otmar Hilliges
Sources and credits
Photo: We have no published photo of Seonwook Park yet. Send us a free-licensed photo.
- Deep Pictorial Gaze Estimation | AIT Lab
- Seonwook Park | AIT Lab
- Seonwook Park | Research
- News | Seonwook (Wookie) Park
- ETH-XGaze: A Large Scale Dataset for Gaze Estimation under Extreme Head Pose and Gaze Variation | AIT Lab
- Lunit Publications | Lunit Research
- GAIA: Generative Animatable Interactive Avatars with Expression-conditioned Gaussians | Research
- [2608.20534] Grounded-Exo2Ego: Structured Semantic Grounding for Robust Exocentric-to-Egocentric Video Generation
- Seonwook (Wookie) Park

