Algorithm 1 rPassD2CGAN calculates each discriminator’s gradient penalty. We use the default values $\lambda =10$, $n_{critic}=10$, $n_{gen}=40$, $\alpha =0.0001$, ${\beta }_1=0.5$ and ${\beta }_2=0.9$.

Require: Gradient penalty coefficient $\lambda$, number of critic iterations per generator $n_{critic}$, number of generator iterations per discriminator $n_{gen}$, batch size m, and Adam hyper-parameters $\alpha$, ${\beta }_1$, and ${\beta }_2$. Require: Initial $D_1$ and $D_2$ critic parameters $w_0$ and $u_0$, and initial generator parameter ${\theta }_0$.  while $\theta$ has not converged do   for  $t=1,\dots ,n_{critic}$  do    for  $i=1,\dots ,m$  do     Sample real data $x\sim {\mathbb{P}}_r$, latent variable $z\sim p\left(z\right)$, and a random number $ϵ\sim U[0,1]$.     $\tilde{x}\leftarrow G_{\theta }\left(z\right)$     $\widehat{x}\leftarrow ϵx+(1-ϵ)\tilde{x}$     $\overline{x}\leftarrow ϵ\tilde{x}+(1-ϵ)x$     $L_{D_1}^i\leftarrow D_w\left(\tilde{x}\right)-D_w\left(x\right)+\lambda {({\parallel {\nabla }_{\widehat{x}}{D}_w\left(\widehat{x}\right)\parallel }_2-1)}^2$     $L_{D_2}^i\leftarrow D_u\left(x\right)-D_u\left(\tilde{x}\right)+\lambda {({\parallel {\nabla }_{\overline{x}}{D}_w\left(\overline{x}\right)\parallel }_2-1)}^2$     $L_{D{2}_{comb}}^i=L_{D_1}^i+L_{D_2}^i$    end for    $(w,u)\leftarrow \mathrm{Adam}({\nabla }_{(w,u)}\frac{1}{m}{\sum }_{i=1}^m{L}_{D{2}_{comb}}^{\left(i\right)},w,u,\alpha ,{\beta }_1,{\beta }_2)$   end for   for  $t=1,\dots ,n_{gen}$  do    Sample a batch of latent variable ${\left\{z^{\left(i\right)}\right\}}_{i=1}^m\sim p_{\left(z\right)}$    $\theta \leftarrow \mathrm{Adam}({\nabla }_{\theta }\frac{1}{m}{\sum }_{i=1}^m(-D_w\left(G_{\theta }\left(z\right)\right)+D_u\left(G_{\theta }\left(z\right)\right),\theta ,\alpha ,{\beta }_1,{\beta }_2)$   end for  end while