A Keras implementation of CapsNet in the paper:
Sara Sabour, Nicholas Frosst, Geoffrey E Hinton. Dynamic Routing Between Capsules. NIPS 2017
This code is adopted from CapsNet-Keras to test the performance of CapsNet on Fashion-MNIST by Xifeng Guo and improved by Norbert Bajko
Contacts
Norbert Bajko
E-mail norbert.bajko94@gmail.com
Xifeng Guo
E-mail guoxifeng1990@163.com or WeChat wenlong-guo.
Step 1. Install Keras with TensorFlow backend
pip install tensorflow-gpu
pip install keras
Step 2. Clone this repository to local
git clone https://github.com/norbertbajko/CapsNet-Fashion-MNIST.git
cd CapsNet-Fashion-MNIST
Step 3. Train a CapsNet on Fashion-MNIST
Training with default settings:
$ python capsulenet.py
Data preprocessing:
- scale pixel values to
[0,1] - shift 2 pixels and horizontal flipping augmentation
Step 4. Test a pre-trained CapsNet model
Suppose you have trained a model using the above command, then the trained model will be
saved to result/trained_model.h5.
Now just launch the following command to get test results:
$ python capsulenet.py --is_training 0 --weights result/trained_model.h5
It will output the testing accuracy and show the reconstructed images. The testing data is same as the validation data. It will be easy to test on new data, just change the code as you want.
Accuracy
Training Speed
About 175s / epoch on a single Titan X (Pascal) GPU.
Reconstruction results
Top 5 rows are real images from Fashion-MNIST and Bottom are corresponding reconstructed images.
