Algorithm 1. The training algorithm for DCGAN-GP.

Input: Number of iterations $N$, discriminator $D$ per iteration, training iterations $k$, batch size $m$, gradient penalty weight $\mathsf{\lambda }$, discriminator update count limit $n_{critic}$ Randomly initialize generator $G$ network parameters ${\mathsf{\theta }}_G$ and discriminator $D$$\mathrm{network} \mathrm{parameters} {\mathsf{\theta }}_G$ $For i=1:N do$ # Train the discriminator 4.$For j=1:k do$ # Collect mini-batch samples 5.$\mathrm{Randomly} \mathrm{sample} \mathrm{m} \mathrm{samples} \{z_1,z_2,\dots ,z_m\}$$\mathrm{from} \mathrm{the} \mathrm{random} \mathrm{noise} \mathrm{distribution} \mathrm{data} p(z)$ 6.Randomly sample $m$$\mathrm{real} \mathrm{samples} \{x_1,x_2,\dots ,x_m\}$$\mathrm{from} \mathrm{the} \mathrm{real} \mathrm{distribution} \mathrm{dataset} p_{data}(x)$ # Compute the random interpolation points for the gradient penalty term. 7.Sample $m$$\mathrm{random} \mathrm{numbers} \{{\mathsf{\lambda }}_1,{\mathsf{\lambda }}_2,\dots ,{\mathsf{\lambda }}_m\}$$\mathrm{from} \mathrm{a} \mathrm{uniform} \mathrm{distribution} \mathrm{in} \mathrm{the} \mathrm{range} [0,1]$ 8.$\mathrm{Construct} \mathrm{interpolation} \mathrm{samples} {\widehat{x}}_i={\mathsf{\lambda }}_i{x}_i+(1-{\mathsf{\lambda }}_i)G(z_i)$ # Calculate the loss function of the discriminator and update the parameters 9.Calculate the loss function of the discriminator 10.Calculate the gradient penalty term 11.$\mathrm{The} \mathrm{gradient} \mathrm{of} \mathrm{the} \mathrm{discriminator} \mathrm{model} \mathrm{parameters} \mathrm{with} \mathrm{respect} \mathrm{to} {\mathsf{\theta }}_D$$\mathrm{is} {ℒ}_D={\mathbb{E}}_{x~P_r}[D(x)]-{\mathbb{E}}_{z~P_z}[D(G(z))]+\mathsf{\lambda }{\mathbb{E}}_{\widehat{x}~P_{\widehat{x}}}[{({‖{\nabla }_{\widehat{x}}D(\widehat{x})‖}_2-1)}^2]$ 12.$End$ # Train the generator 13.Collecting $m$$\mathrm{samples} \{z_1,z_2,\dots ,z_m\}$$\mathrm{from} \mathrm{the} \mathrm{random} \mathrm{noise} \mathrm{distribution} p(z)$ # Calculating the generator’s loss function and updating parameters 14.Calculating the generator’s loss function. 15.$\mathrm{Updating} \mathrm{the} \mathrm{generator} \mathrm{model} \mathrm{parameters} {\mathsf{\theta }}_G$$\mathrm{with} \mathrm{gradient} {\nabla }_{{\mathsf{\theta }}_G}=-{\nabla }_{{\mathsf{\theta }}_G}[D(G(z_i))]$ 16.$End$ 17.Output: Generator $G$