Wouter Van Gansbeke

Staff Research Scientist · Google DeepMind

Hi! I am a Staff Research Scientist at Google DeepMind working on Gemini. My research focuses on large-scale reinforcement learning (RL) for reasoning and autonomous software engineering. I captained major RL post-training runs across several Gemini Thinking revisions and co-led the integration of a new RL algorithm into Gemini 3.0.

Previously, I obtained my PhD in Computer Vision and Machine Learning at KU Leuven advised by Prof. Luc Van Gool, where I worked on self-supervised visual representation learning (SCAN, MaskContrast, ConSSL).

Experience & Education

Google DeepMind – Gemini May 2024 – Present

Staff Research Scientist · London, UK

Uber (Segments.ai, Y Combinator — acquired by Uber) March – Oct. 2023

Research Intern, Computer Vision

Meta – Fundamental AI Research (FAIR) June – Sept. 2022

Research Intern · Menlo Park, USA

KU Leuven 2013 – 2023

PhD in Engineering Science (2018–2023, advisor: Luc Van Gool) · MSc (Cum Laude, 2018) · BSc (2016)

News

Oct. 2026 Released Gemini 4.0, advancing long-horizon reasoning and agentic software engineering.
Nov. 2025 Released Gemini 3.0, integrating our new RL post-training algorithm and reasoning recipe.
June 2025 Technical report on Gemini 2.5 is released.
March 2025 Released Gemini 2.5 Pro with multimodal reasoning (#1 on LMArena).
Dec. 2024 Released experimental Gemini Flash Thinking in Google AI Studio.
May 2024 Joined Google DeepMind in London as a Research Scientist on Gemini.
March 2024 Paper on latent diffusion for panoptic segmentation (LDMSeg) accepted at ECCV 2024.
July 2023 Defended PhD in Engineering Science at KU Leuven and named Outstanding Reviewer at CVPR 2023 (and ECCV 2022).
June 2022 Joined Meta Fundamental AI Research (FAIR) in Menlo Park as a Summer Research Intern.
2020 – 2021 Papers accepted at NeurIPS 2021, ICCV 2021, IEEE TPAMI, and ECCV 2020; co-organized the ICCV 2021 Deep Multi-Task Learning workshop.

Publications

9,000+ citations · * equal contribution

Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities

Gemini Team (including Wouter Van Gansbeke)

arXiv:2507.06261, 2025 [arXiv] [PDF]

@article{gemini2025pushing,
  title={Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities},
  author={Gemini Team},
  journal={arXiv preprint arXiv:2507.06261},
  year={2025}
}

LDMSeg latent diffusion panoptic segmentation teaser

A Simple Latent Diffusion Approach for Panoptic Segmentation and Mask Inpainting

Wouter Van Gansbeke, Bert De Brabandere

ECCV, 2024 [arXiv] [PDF] [Code]

@inproceedings{vangansbeke2024simple,
  title={A Simple Latent Diffusion Approach for Panoptic Segmentation and Mask Inpainting},
  author={Van Gansbeke, Wouter and De Brabandere, Bert},
  booktitle={European Conference on Computer Vision (ECCV)},
  year={2024}
}

Discovering Object Masks with Transformers for Unsupervised Semantic Segmentation

Wouter Van Gansbeke, Simon Vandenhende, Luc Van Gool

arXiv:2206.06363, 2022 [arXiv] [PDF] [Code]
Revisiting Contrastive Methods for Unsupervised Learning of Visual Representations

Wouter Van Gansbeke*, Simon Vandenhende*, Stamatios Georgoulis, Luc Van Gool

NeurIPS, 2021 [arXiv] [PDF] [Code]
Unsupervised Semantic Segmentation by Contrasting Object Mask Proposals

Wouter Van Gansbeke*, Simon Vandenhende*, Stamatios Georgoulis, Luc Van Gool

ICCV, 2021 [arXiv] [PDF] [Code]
Multi-Task Learning for Dense Prediction Tasks: A Survey

Simon Vandenhende, Stamatios Georgoulis, Wouter Van Gansbeke, Marc Proesmans, Dengxin Dai, Luc Van Gool

IEEE TPAMI, 2021 1k+ Citations [arXiv] [PDF] [Code]
SCAN: Learning to Classify Images Without Labels

Wouter Van Gansbeke*, Simon Vandenhende*, Stamatios Georgoulis, Marc Proesmans, Luc Van Gool

ECCV, 2020 1.5k+ GitHub Stars, 1k Citations [arXiv] [PDF] [Code] [Video] [Slides]
Don’t Forget The Past: Recurrent Depth Estimation from Monocular Video

Vaishakh Patil, Wouter Van Gansbeke, Dengxin Dai, Luc Van Gool

IEEE RA-L, 2020 [arXiv] [PDF]
Sparse and Noisy LiDAR Completion with RGB Guidance and Uncertainty

Wouter Van Gansbeke, Davy Neven, Bert De Brabandere, Luc Van Gool

MVA, 2019 [arXiv] [PDF] [Code] [Video]
End-to-End Lane Detection through Differentiable Least-Squares Fitting

Wouter Van Gansbeke, Bert De Brabandere, Davy Neven, Marc Proesmans, Luc Van Gool

ICCV Workshops, 2019 [arXiv] [PDF] [Code]