Figure 2.

The RealNVP neural network implementation of the basic module of the bijector $\psi$ , where $s_1,s_2$ and $t_1,t_2$ are all feed-forward neural networks with three layers, 64 hidden neurons, and ReLU active function. $s_i$ s and $t_i$ s share parameters, respectively. ⨂ and + represent element-wised product and addition, respectively. $\mathbf{x}={\mathbf{x}}_1⨁{\mathbf{x}}_2$ and ${\mathbf{x}}^{\prime }={\mathbf{x}}_1^{\prime }⨁{\mathbf{x}}_2^{\prime }$ .