An experimental Generative Adversarial Network implementation of the Capsule Neural Network. It is based on the recent works on Dynamic Routing between Capsules: reference: Dynamic routing between capsules by Sara Sabour, Nicholas Frosst, Geoffrey E Hinton
- Tensorflow
- Numpy
- Pandas
- DateTime
- Matplotlib
pip install -r requirements.txt
python CapsNetGANs.py
Currently the initial commit is implemented to train on the domains of the MNIST Dataset.
Resulting generated images of the network are on the way. I'll soon commit them.
-If the CapsNetGANs.py does not work on your machine due to GPU Memory constraint, try CapsNetGANs2.py, in this only the discriminator has the CapsNet layer and thus takes much lesser memory.
-Added the CapsNet layer to the generative model, but the entire network hogs a lot of GPU memory, need to find a way to reduce it in the next commit. Optimised for memory constraints.
-Need to tune hyperparameter so that the GAN gets properly trained.
-Add an option to also train on Fashion MNIST dataset
- https://github.com/llSourcell/Generative_Adversarial_networks_LIVE For EZGAN implementation
- https://github.com/naturomics/CapsNet-Tensorflow For CapsNet Implementation