I have used an inverted CNN as a generator and CNN as a Discriminator .Firstly,Trained the discriminator using real images , then with fake images through the 3 channels provided by the generator.Used ReLU activation for forward propogation in Generator network, LeakyReLU for forward propogation in Discriminator network and then computed the losses using binary cross entropy and updated the weights through backpropogation .
Check Results folder for generated images .