[64 channel 5x5 CONV + ReLU] x5
[3 channel 3x3 CONV]
And for the discriminator:
[64 channel 3x3 CONV + ReLU]
[64 channel 3x3 CONV stride 2 + ReLU]
[64 channel 3x3 CONV stride 1 + ReLU]
[64 channel 3x3 CONV stride 2 + ReLU]
[64 channel 1x1 CONV stride 1]
[Spatial Global Average Pooling]
[Sigmoid]