Bispectral Neural Networks

This repository is the official implementation of Bispectral Neural Networks.

Installation

To install the requirements and package, run:

pip install -r requirements.txt
python setup.py install

Datasets

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 'https://docs.google.com/uc?export=download&id=10w3fKdO0eWEe2KxZxpf8YFndXdCNNR8b' -O- | sed -rn 's/.*confirm=([0-9A-Za-z_]+).*/\1\n/p')&id=10w3fKdO0eWEe2KxZxpf8YFndXdCNNR8b" -O datasets.zip 
rm -rf /tmp/cookies.txt
unzip datasets.zip
rm -r datasets.zip

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

1. Download the zip file here.
2. Place the file in the top node of this directory, i.e. in bispectral-networks/.
3. Run:

   unzip datasets.zip
   rm -r datasets.zip
   

Training

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

python train.py --config rotation_experiment
python 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 the configs/ folder.

To view learning curves in Tensorboard, run:

tensorboard --logdir logs/

Pre-trained Models

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

logs/rotation_model/
logs/translation_model/

Results and Figures

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

notebooks/rotation_experiment_analysis.ipynb
notebooks/translation_experiment_analysis.ipynb

License

This repository is licensed under the MIT License.