def discriminator(x, f_dim, output_size, c_dim, is_training=True): # Network net = x with slim.arg_scope([conv2d], activation_fn=lrelu, is_spectral_norm=True, kernel_size=[5, 5], stride=[2, 2], update_collection=None): net = conv2d(net, f_dim, name='conv0') net = conv2d(net, f_dim, name='conv1') net = conv2d(net, f_dim, name='conv2') net = conv2d(net, f_dim, name='conv3') net = tf.reshape(net, [-1, output_size // 16 * output_size // 16 * f_dim]) logits = fully_connected(net, 1, activation_fn=None, is_spectral_norm=True, update_collection=None, name='fc4') logits = tf.squeeze(logits, -1) return logits
def discriminator(x, f_dim, output_size, c_dim, is_training=True): # Network net = x with slim.arg_scope([conv2d], activation_fn=lrelu, is_spectral_norm=False, update_collection=None): net = conv2d(net, f_dim, kernel_size=[3, 3], stride=[1, 1], name='conv0_0') net = conv2d(net, f_dim, kernel_size=[4, 4], stride=[2, 2], name='conv0_1') net = conv2d(net, 2 * f_dim, kernel_size=[3, 3], stride=[1, 1], name='conv1_0') net = conv2d(net, 2 * f_dim, kernel_size=[4, 4], stride=[2, 2], name='conv1_1') net = conv2d(net, 4 * f_dim, kernel_size=[3, 3], stride=[1, 1], name='conv2_0') net = conv2d(net, 4 * f_dim, kernel_size=[4, 4], stride=[2, 2], name='conv2_1') net = conv2d(net, 8 * f_dim, kernel_size=[3, 3], stride=[1, 1], name='conv3_0') net = tf.reshape(net, [-1, output_size // 8 * output_size // 8 * 8 * f_dim]) # net = slim.fully_connected(net, 512, activation_fn=lrelu) logits = fully_connected(net, 1, activation_fn=None, is_spectral_norm=False, update_collection=None) logits = tf.squeeze(logits, -1) return logits
def discriminator(x, f_dim, output_size, c_dim, is_training=True): # Network net = x argscope_conv2d = slim.arg_scope([conv2d], kernel_size=[4, 4], stride=[2, 2], activation_fn=tf.nn.relu, is_spectral_norm=True, update_collection=None) with argscope_conv2d: net = conv2d(net, f_dim, name='conv0_0') dnet = conv2d(net, f_dim, kernel_size=[3, 3], stride=[1, 1], name='conv0_1') net += 1e-1 * conv2d(dnet, f_dim, kernel_size=[3, 3], stride=[1, 1], name='conv0_2') net = conv2d(net, f_dim, name='conv1_0') dnet = conv2d(net, f_dim // 2, kernel_size=[3, 3], stride=[1, 1], name='conv1_1') net += 1e-1 * conv2d(dnet, f_dim, kernel_size=[3, 3], stride=[1, 1], name='conv1_2') net = conv2d(net, f_dim, name='conv2_0') dnet = conv2d(net, f_dim // 2, kernel_size=[3, 3], stride=[1, 1], name='conv2_1') net += 1e-1 * conv2d(dnet, f_dim, kernel_size=[3, 3], stride=[1, 1], name='conv2_2') net = conv2d(net, f_dim, name='conv3_0') dnet = conv2d(net, f_dim // 2, kernel_size=[3, 3], stride=[1, 1], name='conv3_1') net += 1e-1 * conv2d(dnet, f_dim, kernel_size=[3, 3], stride=[1, 1], name='conv3_2') net = tf.reshape(net, [-1, output_size // 16 * output_size // 16 * f_dim]) logits = fully_connected(net, 1, activation_fn=None, is_spectral_norm=True, update_collection=None, name='fc4') logits = tf.squeeze(logits, -1) return logits