A DCGAN is simply a GAN that uses a convolutional neural network as the discriminator, and a network composed of transposed convolutions as the generator. We will be using this to generate new emojis from a dataset of emojis.
We implement the following to improve output.
- Reduced learning rate
- Normalized inputs
- Applied smoothing to real and fake labels
- Used normally distributed sample noise
- Used custom kernel initialization
- Added batch normalization and leaky relu at the output of each convolutional layer in discriminator
The new emojis are shown below.
