CapsuleNet-Gluon

Implemention of Capsule Net from the paper Dynamic Routing Between Capsules Edit Add topics

Run

1. pip install --pre mxnet-cu80 -i https://pypi.douban.com/simple --user
2. pip install tqdm
3. python main.py

Results

Results to be added

Issues

I use SGD to try Capsule Net, if use Adam as described in the original paper, I got very low test accuracy.

I find if use large train batch_size, the results is also unaceptable. I haven't figure it out.

Any PR is welcomed.

Other Implementations

• Kaggle (this version as self-contained notebook):

  • MNIST Dataset running on the standard MNIST and predicting for test data
  • MNIST Fashion running on the more challenging Fashion images.

• TensorFlow:

  • naturomics/CapsNet-Tensorflow
    Very good implementation. I referred to this repository in my code.
  • InnerPeace-Wu/CapsNet-tensorflow
    I referred to the use of tf.scan when optimizing my CapsuleLayer.
  • LaoDar/tf_CapsNet_simple

• PyTorch:

  • tonysy/CapsuleNet-PyTorch
  • timomernick/pytorch-capsule
  • gram-ai/capsule-networks
  • andreaazzini/capsnet.pytorch
  • leftthomas/CapsNet

• MXNet:

  • AaronLeong/CapsNet_Mxnet

• Lasagne (Theano):

  • DeniskaMazur/CapsNet-Lasagne

• Chainer:

  • soskek/dynamic_routing_between_capsules