Robert Geirhos
Staff Research Scientist at Google DeepMind
Zurich, Zurich, Switzerland
Summary
👤
Senior🎓
Top SchoolRobert Geirhos is a senior research scientist with nearly a decade of focused expertise at the intersection of deep learning and visual neuroscience, currently leading work at Google DeepMind. His research—recognized with awards and influential publications such as the ICLR 2019 oral paper on texture vs. shape bias—bridges rigorous human vision experiments and practical model improvements that enhance robustness and generalization. He built widely used assets, including pre-trained models and analysis tooling for the texture-vs-shape benchmark, reflecting both strong experimental and engineering skills. A summa cum laude doctoral graduate from the University of Tübingen, he combines academic rigor with impact-driven industry research across Google Brain, FAIR, and top labs in Europe. Colleagues cite his knack for translating cognitive insights into concrete model interventions that improve performance in real-world vision tasks.
9 years of coding experience
4 years of employment as a software developer
Exchange student, Computer Science - Psychology - Statistics, Exchange student, Computer Science - Psychology - Statistics at University of GlasgowMaster’s Degree, Computer Science (with distinction), Master’s Degree, Computer Science (with distinction) at University of TübingenExchange student, Computer Science, Exchange student, Computer Science at University of AmsterdamDoctor of Science, Computer Science, Doctor of Science, Computer Science at University of Tübingen & International Max Planck Resarch School for Intelligent Systems
Github Skills (13)
data-loading10
computer-vision10
machine-learning10
preloading10
deep-learning10
resource-loading10
r10
data-analysis10
shapes9
python9
texture9
pytorch8
tensorflow6
Programming languages (8)
Github contributions (5)
rgeirhos/texture-vs-shape
Nov 2018 - Mar 2022
Pre-trained models, data, code & materials from the paper "ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness" (ICLR 2019 Oral)
Role in this project:
Data Scientist
Data ScientistContributions:41 commits, 32 pushes, 42 comments in 3 years 4 months
Contributions summary:Robert primarily contributed to the data analysis aspects of the repository, adding a helper script (`data-analysis-helper.R`) containing numerous functions for data analysis related to texture-vs-shape experiments. Further updates involved removing obsolete plotting functionalities from the main `data-analysis.R` script. Additionally, the user added and updated a model loading file (`load_pretrained_models.py`) which included the definition of various ResNet50 models trained on different datasets, showcasing an understanding of model loading and potentially model evaluation. The user added and updated a model loading file (`load_pretrained_models.py`) which included the definition of various AlexNet and VGG16 models trained on different datasets.
imagenetpre-trained-modeldeep-learningpsychophysicshuman-vision
google-deepmind/physics-IQ-benchmark
Jan 2025 - Jun 2026
Benchmarking physical understanding in generative video models
Contributions:14 reviews, 18 PRs, 53 pushes in 1 year 5 months
benchmarkingbenchmarkgenerative-modelsvideo-generationphysical-understanding
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.