This repository is the official accompaniment toA General Framework for Robust G-Invariance in G-Equivariant Networks (submitted, NeurIPS 2023)

To install the requirements and package, run:

pip install -r requirements.txtpython install -e .

To download the datasets, run:

wget --load-cookies /tmp/cookies.txt "https://docs.google.com/uc?export=download&confirm=$(wget --quiet --save-cookies /tmp/cookies.txt --keep-session-cookies --no-check-certificate '[URL]' -O- | sed -rn 's/.*confirm=([0-9A-Za-z_]+).*/\1\n/p')&id=[ID]" -O datasets.zip rm -rf /tmp/cookies.txtunzip datasets.ziprm -r datasets.zip

If your machine doesn't have wget, follow these steps:

1. Download the zip filehere.
2. Place the file in the top node of this directory, i.e. intc-invariance/.
3. Run:

   unzip datasets.ziprm -r datasets.zip

To train the models in the paper, run the following commands.

python train.py --config rotation_experimentpython train.py --config translation_experiment

To run on GPU, add the following argument, with the integer specifying the device number, i.e.:

--device 0

The full set of hyperparameters and training configurations are specified in the config files in theconfigs/ folder.

To view learning curves in Tensorboard, run:

tensorboard --logdir logs/

The pre-trained models are included in the repo, in the following locations:

logs/gtc/logs/max/

All results and figures from the paper are generated in the Jupyter notebooks located at:

notebooks/file.ipynb

This repository is licensed under the MIT License.

PinnedLoading

1. gtc-invariancegtc-invariancePublic

   This repository is the official accompaniment to A General Framework for Robust G-Invariance in G-Equivariant Networks (submitted, NeurIPS 2023)

   3 2