The below GIF displays the sample of images generated from epoch 1 to 50 at every 5 epochs.
Conv layers enable GANs to generate better images much faster than neural net.
Each epoch takes around 60 seconds
The below GIF displays the sample of images generated from epoch 1 to 200 at every 20 epochs.
Neural net enables GANs to generate decent images but after much longer training epochs.
Each epoch takes around 15 seconds.
The below GIF displays the sample of images generated from epoch 1 to 9 at every epoch.
At the decoder end a 28x28 image is reconstructed by passing the latent vector along with its true class variable through two fully connected layers
Each epoch takes around 55 mins seconds.
- Tensorflow
- Keras
- openCV
- PIL
- numpy
- GANs, https://papers.nips.cc/paper/5423-generative-adversarial-nets.pdf
- https://www.oreilly.com/learning/generative-adversarial-networks-for-beginners
- https://medium.com/@devnag/generative-adversarial-networks-gans-in-50-lines-of-code-pytorch-e81b79659e3f
- https://jhui.github.io/2017/11/03/Dynamic-Routing-Between-Capsules/
- https://kndrck.co/posts/capsule_networks_explained/
- https://ctmakro.github.io/site/on_learning/fast_gan_in_keras.html
- Overview of GANs, https://arxiv.org/pdf/1710.07035.pdf
- Capsule Nets, https://arxiv.org/pdf/1710.09829.pdf
- https://github.com/XifengGuo/CapsNet-Keras