Generater model : structure of generator model is taken from UWGAN
here we used skip connections and msdb layers to improve performance of our model.
Discriminator model : For discriminator model I have used the inception model cause it is very good at discriminating images even if they are almost same.
Loss function :
loss = Lgan (D, G) + λ1 * L1(G)
Lgan = Ex [log D(x)] + Ey[log(1 − D(G(y)))]
L1 is abslute distance between the original image and the generated image.
λ1 is set to 50
Download and run GAN.ipynb file in google colab