def __init__(self, in_size, out_size, decay=0.9, eps=2e-5, dtype=numpy.float32):
super(SharedEmbeddingBatchNormalization, self).__init__(
size=out_size, decay=decay, eps=eps, dtype=dtype)
with self.init_scope():
self.linear_gamma = SNLinear(in_size, out_size, nobias=True)
self.linear_beta = SNLinear(in_size, out_size, nobias=True)
def __init__(self, ch=64, n_classes=0, activation=F.relu):
super(SNResNetProjectionDiscriminator, self).__init__()
self.activation = activation
initializer = chainer.initializers.GlorotUniform()
with self.init_scope():
self.block1 = OptimizedBlock(3, ch)
self.block2 = Block(ch,
ch * 2,
activation=activation,
downsample=True)
self.block3 = Block(ch * 2,
ch * 4,
activation=activation,
downsample=True)
self.block4 = Block(ch * 4,
ch * 8,
activation=activation,
downsample=True)
self.block5 = Block(ch * 8,
ch * 16,
activation=activation,
downsample=True)
self.block6 = Block(ch * 16,
ch * 16,
activation=activation,
downsample=False)
self.l7 = SNLinear(ch * 16, 1, initialW=initializer)
if n_classes > 0:
self.l_y = SNEmbedID(n_classes, ch * 16, initialW=initializer)
def __init__(self, ch=128, n_classes=0, activation=F.relu):
super(SNResNetProjectionDiscriminator, self).__init__()
self.activation = activation
with self.init_scope():
self.block1 = OptimizedBlock(3, ch)
self.block2 = Block(ch, ch, activation=activation, downsample=True)
self.block3 = Block(ch,
ch,
activation=activation,
downsample=False)
self.block4 = Block(ch,
ch,
activation=activation,
downsample=False)
self.l5 = SNLinear(ch,
1,
initialW=chainer.initializers.GlorotUniform(),
nobias=True)
self.l5.u = np.ones((1, 1)).astype(np.float32)
if n_classes > 0:
self.l_y = SNEmbedID(
n_classes,
ch,
initialW=chainer.initializers.GlorotUniform())
seed_weights(self)
def __init__(self,
in_size,
out_size,
decay=0.9,
eps=2e-5,
dtype=numpy.float32,
use_moving_average=True):
super(HyperBatchNormalization,
self).__init__(size=out_size,
decay=decay,
eps=eps,
dtype=dtype,
use_moving_average=use_moving_average)
with self.init_scope():
self.linear_gamma = SNLinear(in_size, out_size, nobias=True)
self.linear_beta = SNLinear(in_size, out_size, nobias=True)
def __init__(self, ch=128, n_classes=0, activation=F.relu, sn=False, ch_input=3):
super(SNResNetProjectionDiscriminator, self).__init__()
self.activation = activation
self.n_classes = n_classes
with self.init_scope():
self.block1 = OptimizedBlock(ch_input, ch, sn=sn)
self.block2 = Block(ch, ch, activation=activation, downsample=True, sn=sn)
self.block3 = Block(ch, ch, activation=activation, downsample=False, sn=sn)
self.block4 = Block(ch, ch, activation=activation, downsample=False, sn=sn)
self.l5 = SNLinear(ch, 1, initialW=chainer.initializers.GlorotUniform(), nobias=True)
if n_classes > 0:
self.l_y = SNEmbedID(n_classes, ch, initialW=chainer.initializers.GlorotUniform())
def __init__(self, ch=96):
self.ch = ch
super(Generator, self).__init__()
with self.init_scope():
self.linear = L.Linear(1000, 128, nobias=True)
self.G_linear = SNLinear(20, 4 * 4 * 16 * ch)
self.GBlock = ResBlock(148, 16 * ch, 16 * ch)
self.GBlock_1 = ResBlock(148, 16 * ch, 8 * ch)
self.GBlock_2 = ResBlock(148, 8 * ch, 8 * ch)
self.GBlock_3 = ResBlock(148, 8 * ch, 4 * ch)
self.GBlock_4 = ResBlock(148, 4 * ch, 2 * ch)
self.attention = NonLocalBlock(2 * ch)
self.GBlock_5 = ResBlock(148, 2 * ch, ch)
self.ScaledCrossReplicaBN = L.BatchNormalization(ch)
self.conv_2d = SNConvolution2D(ch, 3, 3, 1, 1)
def __init__(self, ch=128, n_classes=0, activation=F.relu):
super(SRDiscriminator, self).__init__()
self.activation = activation
with self.init_scope():
self.block1 = OptimizedBlock(3, ch)
self.block2 = Block(ch, ch, activation=activation, downsample=True)
self.block3 = Block(ch,
ch,
activation=activation,
downsample=False)
self.block4 = Block(ch,
ch,
activation=activation,
downsample=False)
self.l5 = SNLinear(ch,
1,
initialW=chainer.initializers.GlorotUniform(),
nobias=True)
if n_classes > 0:
self.l_y = SNEmbedID(
n_classes,
ch,
initialW=chainer.initializers.GlorotUniform())
