Installation:

pip install -e .

Installing DL2:

First clone DL2 in a separate directory and install it using the following commands:

git clone https://github.com/eth-sri/dl2
cd dl2
pip install -r requirements.txt

If you are using a virtual environment then make sure to install DL2 in that environment. Now DL2 can be imported as a python libary. To achieve this just extend the PYTHONPATH variable to also point to the DL2 directory:

export PYTHONPATH="${PYTHONPATH}:{path_to_dl2}"

Execution:

Run CIFAR10 experiment:

run_image_experiments.py cifar10 --layers=10 --widen_factor=1 run

Run CIFAR100 experiment:

run_image_experiments.py cifar100 --layers=10 --widen_factor=1 run

Generate experiment conditions:

cd scripts
python generate_experiments.py

Help functions:

run_image_experiments.py -- --help
run_image_experiments.py cifar10 -- --help
run_image_experiments.py cifar100 -- --help

Acknowledgements

• densenet-pytorch
• Wide Residual Networks (WideResNets) in PyTorch
• Wide Residual Networks (BMVC 2016) http://arxiv.org/abs/1605.07146 by Sergey Zagoruyko and Nikos Komodakis.