This is a CapsNet repo which is a tenserflow version for Dynamic Routing Between Capsules(Geoffrey E. Hinton et al)
- Picture below shows the framework of CapsNet, more details can be seen in this repo.
- Picture below shows the reconstruction processing, more details can be seen in this repo.
Picture below shows how routing work, more details can be seen in this repo.
tensorflow 1.3.0
python 2.7.12
numpy 1.13.1
scipy 0.17.0
(1)download this repo to your own directory
$ git clone https://github.com/nnUyi/CapsNet.git
(2)download mnist dataset and store it in the data directory(directory named data)
(3)training
$ python main.py --is_training=True --mask_with_y=True
(4)testing
$ python main.py --is_training=False --mask_with_y=False
The result shows below, after 15 epoches, we can clearly obtain such a good experimental result.
This repo is finished by referring to naturomics-CapsNet-Tensorflow
Email:computerscienceyyz@163.com, Thank you for contacting if you find something wrong or if you have some problems!!!