def __init__(self,
in_channels,
out_channels,
ksize=3,
pad=1,
activation=F.relu):
super(OptimizedBlock, self).__init__()
initializer = chainer.initializers.GlorotUniform(math.sqrt(2))
initializer_sc = chainer.initializers.GlorotUniform()
self.activation = activation
with self.init_scope():
self.c1 = SNConvolution2D(in_channels,
out_channels,
ksize=ksize,
pad=pad,
initialW=initializer)
self.c2 = SNConvolution2D(out_channels,
out_channels,
ksize=ksize,
pad=pad,
initialW=initializer)
self.c_sc = SNConvolution2D(in_channels,
out_channels,
ksize=1,
pad=0,
initialW=initializer_sc)
def __init__(self,
in_channels,
out_channels,
hidden_channels=None,
ksize=3,
pad=1,
activation=F.relu,
downsample=False):
super(Block, self).__init__()
initializer = chainer.initializers.GlorotUniform(math.sqrt(2))
initializer_sc = chainer.initializers.GlorotUniform()
self.activation = activation
self.downsample = downsample
hidden_channels = in_channels if hidden_channels is None else hidden_channels
with self.init_scope():
self.c1 = SNConvolution2D(in_channels,
hidden_channels,
ksize=ksize,
pad=pad,
initialW=initializer)
self.c2 = SNConvolution2D(hidden_channels,
out_channels,
ksize=ksize,
pad=pad,
initialW=initializer)
self.c_sc = SNConvolution2D(in_channels,
out_channels,
ksize=1,
pad=0,
initialW=initializer_sc)
def __init__(self, in_size, in_channel, out_channel):
super(ResBlock, self).__init__()
with self.init_scope():
self.HyperBN = HyperBatchNormalization(in_size, in_channel)
self.conv0 = SNConvolution2D(in_channel, out_channel, 3, 1, 1)
self.HyperBN_1 = HyperBatchNormalization(in_size, out_channel)
self.conv1 = SNConvolution2D(out_channel, out_channel, 3, 1, 1)
self.conv_sc = SNConvolution2D(in_channel, out_channel, 1, 1, 0)
def __init__(self, ch):
self.ch = ch
super(NonLocalBlock, self).__init__()
with self.init_scope():
self.theta = SNConvolution2D(ch, ch // 8, 1, 1, 0, nobias=True)
self.phi = SNConvolution2D(ch, ch // 8, 1, 1, 0, nobias=True)
self.g = SNConvolution2D(ch, ch // 2, 1, 1, 0, nobias=True)
self.o_conv = SNConvolution2D(ch // 2, ch, 1, 1, 0, nobias=True)
self.gamma = L.Parameter(np.array(0, dtype="float32"))
def __init__(self, in_channels, out_channels, hidden_channels=None, ksize=3, pad=1,
activation=F.relu, downsample=False):
super(Block, self).__init__()
initializer = chainer.initializers.GlorotUniform(math.sqrt(2))
initializer_sc = chainer.initializers.GlorotUniform()
self.activation = activation
self.downsample = downsample
self.learnable_sc = (in_channels != out_channels) or downsample
hidden_channels = in_channels if hidden_channels is None else hidden_channels
with self.init_scope():
self.c1 = SNConvolution2D(in_channels, hidden_channels, ksize=ksize, pad=pad, initialW=initializer)
self.c2 = SNConvolution2D(hidden_channels, out_channels, ksize=ksize, pad=pad, initialW=initializer)
if self.learnable_sc:
self.c_sc = SNConvolution2D(in_channels, out_channels, ksize=1, pad=0, initialW=initializer_sc)
self.c_sc.u = np.ones((1, out_channels)).astype(np.float32)
self.c1.u = np.ones((1, 128)).astype(np.float32)
self.c2.u = np.ones((1, 128)).astype(np.float32)
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)