Wouter Van Gansbeke
Senior Research Scientist at Google DeepMind
London, England, United Kingdom
Summary
🤩
Rockstar🎓
Top SchoolWouter Van Gansbeke is a Senior Research Scientist at Google DeepMind in London with nine years of experience at the intersection of large-scale multimodal foundation models and reinforcement learning for reasoning. He led and co-captained extensive RL runs that contributed to Gemini 3.0 and has focused post-training work to boost Gemini’s reasoning and visual perception (Gemini Thinking). His PhD from KU Leuven under Luc Van Gool centered on self-supervised learning for visual scene understanding, and he has hands-on experience deploying unsupervised and end-to-end vision systems—evidenced by contributions to notable repos like SCAN (unsupervised classification) and lane detection for autonomous driving. He combines deep research rigor with practical engineering, from integrating training pipelines to refining loss functions and evaluation scripts. Outside work he’s an avid hiker and cyclist, and he maintains an active GitHub and personal website that surface his reproducible code and experiments.
9 years of coding experience
6 years of employment as a software developer
Master of Science in Engineering Electrical Engineering with a focus on Information Technology, Master of Science in Engineering Electrical Engineering with a focus on Information Technology at KU Leuven
Github Skills (20)
pytorch10
python10
image-classification10
data-loading10
deeplearning-ai10
unsupervised-learning10
deep-learning10
self-supervised-learning10
computer-vision10
architecture9
machine-learning9
contrastive-learning9
architectures9
architect9
clustering9
Programming languages (2)
Github contributions (5)
wvangansbeke/LaneDetection_End2End
Jan 2019 - May 2020
End-to-end Lane Detection for Self-Driving Cars (ICCV 2019 Workshop)
Role in this project:
ML Engineer
ML EngineerContributions:37 commits, 30 pushes, 125 comments in 1 year 3 months
Contributions summary:Wouter made several updates to the `lanedetection_end2end` repository, focusing on the dataloader, backprojection loss, and Birds Eye View loss components. These changes included modifications to data loading, model architectures, and result writing processes. The commits suggest the user was involved in refining the end-to-end lane detection pipeline, likely optimizing data handling and potentially improving the model's performance. The edits to the loss functions indicate a focus on training and evaluation aspects of the lane detection model.
pytorchiccvlane-detectioniccv-2019end-to-end
wvangansbeke/Unsupervised-Classification
Jul 2020 - Jun 2022
SCAN: Learning to Classify Images without Labels, incl. SimCLR. [ECCV 2020]
Role in this project:ML Engineer
ML EngineerContributions:42 commits, 4 PRs, 47 pushes in 1 year 11 months
Contributions summary:Wouter primarily contributed to the project by modifying and adding to scripts related to model evaluation, self-labeling, and the main SCAN training process. They updated configuration files, integrated new dependencies, and also added a tutorial file to explain and guide the user. Their contributions demonstrate a focus on the implementation, training, and evaluation of unsupervised image classification models.
pytorcheccv2020scansimclrunsupervised-learning
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.