Fig. 2.
Network architecture (a) and differences between the Markovian and the traditional discriminators (b, c). Conv convolution operation, DeConv transposed convolution operation, kN kernel size, sN stride size, cN number of output channels, BN batch normalization layers, ReLU rectified linear unit, LReLU, Leaky rectified linear unit. The green dotted line in generator represents skip connection followed by concatenation operation. pij (i = 1,2,…,n; j = 1,2,…,n) denotes the discriminate probability computed in each local region, where i and j are the orders of regions in the horizontal and the vertical orientations