
Academic lab · Tubingen, Germany
Bethge LabTuebingen
Tübingen lab studying robustness and human-like vision, known for the Foolbox toolbox and open robustness benchmarks.
Tubingen, Germany
- Founded
- 2014
- Type
- Academic lab
- Based in
- Tubingen, Germany
- Key papers
- 3
- Featured projects
- 8
About
The Bethge Lab is led by Matthias Bethge, Professor of Computational Neuroscience and Machine Learning at the University of Tübingen and director of the Tübingen AI Center. Its GitHub organization, created in 2014, carries the tagline Perceiving Neural Networks.
The lab released Foolbox in 2017, a Python toolbox for benchmarking adversarial robustness under an MIT licence, and imagecorruptions and a detection robustness benchmark in 2019. It also published work on ImageNet-C robustness and on comparing machine and human vision.
Wieland Brendel, Jonas Rauber and Robert Geirhos appear among its public GitHub members and papers, and its code serves vision and robustness researchers.
Mission and ambitionsThe lab seeks to understand how neural networks perceive images and to make them robust and more human-like, releasing benchmarks and tools openly.
Known for
- 01
Foolbox adversarial toolbox
- 02
Corruption robustness benchmarks
- 03
Human versus machine vision
- 04
ImageNet-C robustness training
Milestones
- 2017
Foolbox
Released an adversarial robustness toolbox (arXiv 1707.04131).
- 2019
Robustness benchmarks
Released the imagecorruptions package and a detection robustness benchmark (arXiv 1907.07484).
- 2020
Noise training baseline
Rusak et al. showed Gaussian and Speckle noise training is a strong ImageNet-C baseline (arXiv 2001.06057).
- 2021
Human versus machine vision
Released an open benchmark comparing models to humans (arXiv 2106.07411).
Open source
A Python toolbox to create adversarial examples that fool neural networks in PyTorch, TensorFlow, and JAX (2,984 stars)
Python package to corrupt arbitrary images. (486 stars)
Benchmark your model on out-of-distribution datasets with carefully collected human comparison data (NeurIPS 2 (365 stars)
Siamese Mask R-CNN model for one-shot instance segmentation (354 stars)
Model with 1 likes on the Hub.
Model with 0 likes on the Hub.
- bethgelab/CuratedThoughtsdataset
Dataset with 45 likes on the Hub.
Space with 17 likes on the Hub.
Research papers
Notable releases
Openness
Code and datasets are open on GitHub and Hugging Face.
People
- Matthias BethgeProfessor and lab head, director of the Tübingen AI Center
- Wieland BrendelResearcher, Foolbox and robustness papers
- Jonas RauberFoolbox lead author
- Robert GeirhosResearcher on shape and texture bias
- Claudio MichaelisDetection robustness benchmark author
Listed from the lab's own pages and public profiles.
Contact and community
Contact via site.
Sources
Compiled from the lab's public pages, Hugging Face and GitHub. Last checked 2026-11-17. Something wrong or out of date? Tell us on Discord.



