THE CLEVER HANS EFFECT IN UNSUPERVISED LEARNING

Authors:

• Jacob Kauffmann
• Jonas Dippel
• Lukas Ruff
• Wojciech Samek
• Gregoire Montavon
• Klaus-Robert Müller

Usage:

1. Data Preparation

Download the required datasets:

• NIH CXR8: https://academictorrents.com/details/e615d3aebce373f1dc8bd9d11064da55bdadede0
• GitHub COVID-19 image collection: https://github.com/ieee8023/covid-chestxray-dataset
• ImageNet: https://www.image-net.org/download.php
• MVTec-AD: https://www.mvtec.com/company/research/datasets/mvtec-ad

Place the datasets in the data/ directory.

2. Running the experiments

2.1 Software dependencies

• All experiments are implemented in Python.
• Main dependencies: torch, torchvision, ipython, matplotlib, scikit-learn, scipy, numpy, Pillow, opencv-python-headless.
• Experiments provide individual requirements.txt files.

2.2 COVID-19

• Navigate to the radiology/ directory.
• Open covid19.ipynb with ipython.
• Run all cells.
• Results can be found in the cell outputs and the results/ sub-directory.

2.3 Anomaly Detection

• Additional dependency: Snakemake.
• Navigate to the anomalies/ directory.
• Run snakemake --cores 12.
• Results can be found in the results/ sub-directory.

2.4 Representation Learning

• Navigate to the representation/ directory.
• Follow instructions in the README.md file in the respective folder.

Contact Information:

• Grégoire Montavon: gregoire.montavon@fu-berlin.de
• Klaus-Robert Müller: klaus-robert.mueller@tu-berlin.de