Playground for experimentation with CapsNet.

To initiate the training, use the command:

python capsnet.py -t -s -c -v --batchSize 500 --trainingEpochs 100 -d MNIST

where -t stands for training, -s for training from scratch, -c stands for model testing after training, -v is used for tensorboard output, and -d specifies the dataset to be used. The system supports two datasets at this point (MNIST and CIFAR-10). Keep in mind that enabling the tensorboard visualization will take up much more GPU memory, hence, might require you to reduce the batch size.

For just testing a trained model, use the command:

python capsnet.py -c --batchSize 500 -d MNIST

where absence of -t flag indicates that only testing of the trained model has to be performed.

To try CapsNet without routing by agreement (only capsules), execute the capsnet_no_rba.py script as follows:

python capsnet_no_rba.py -t -s -c -v --batchSize 500 --trainingEpochs 20 -d MNIST

1. https://github.com/ageron/handson-ml/blob/master/extra_capsnets.ipynb

Author: Shoaib Ahmed Siddiqui
Email: [email protected]