cyclegan_gen.py

# Generator
def generator(inputgen, name="generator", reuse=None):
    if reuse:
        scope.reuse_variables()

    with tf.variable_scope(name):
        f = 7
        ks = 3

        pad_input = tf.pad(inputgen,[[0, 0], [ks, ks], [ks, ks], [0, 0]], "REFLECT")
        x = conv2d(pad_input, ngf, f, f, 1, 1, 0.02,name="c1")
        x = conv2d(x, ngf*2, ks, ks, 2, 2, 0.02,"SAME","c2")
        x = conv2d(x, ngf*4, ks, ks, 2, 2, 0.02,"SAME","c3")

        x = resnet_block(x, ngf*4, "r1")
        x = resnet_block(x, ngf*4, "r2")
        x = resnet_block(x, ngf*4, "r3")
        x = resnet_block(x, ngf*4, "r4")
        x = resnet_block(x, ngf*4, "r5")
        x = resnet_block(x, ngf*4, "r6")
        x = resnet_block(x, ngf*4, "r7")
        x = resnet_block(x, ngf*4, "r8")
        x = resnet_block(x, ngf*4, "r9")

        x = general_deconv2d(x, [batch_size,128,128,ngf*2], ngf*2, ks, ks, 2, 2, 0.02,"SAME","c4")
        x = general_deconv2d(x, [batch_size,256,256,ngf], ngf, ks, ks, 2, 2, 0.02,"SAME","c5")
        x = conv2d(x, img_layer, f, f, 1, 1, 0.02,"SAME","c6",do_relu=False)

        # Adding the tanh layer
        x = tf.nn.tanh(x,"t1")

        return x