Deep Convolutional Generative Adverasrial Network in Keras (Tensorflow backend)
A DCGAN model to generate images from the MNIST handwritten digit dataset .
1 . Images are normalized to [-1,1]
2 . Activation function for the last layer - tanh
3 . Activation function for other layers - LeakyReLU
4 . No Fully Connected Layers used in the model .
5 . Upsampling followed by Convolution layers are used to increase the resolution of the image .
6 . Optimizer : Adam
Result after 80 epochs :