def initialize(self,
size,
outchn,
dilation_rate=1,
stride=1,
pad='SAME',
activation=-1,
batch_norm=False,
usebias=True,
kernel_data=None,
bias_data=None,
weight_norm=False):
self.conv = L.conv1D(size,
outchn,
stride=stride,
pad=pad,
usebias=usebias,
kernel_data=kernel_data,
bias_data=bias_data,
dilation_rate=dilation_rate,
weight_norm=weight_norm)
self.batch_norm = batch_norm
self.activation_ = activation
if batch_norm:
self.bn = L.batch_norm()
if activation != -1:
self.activation = L.activation(activation)
def initialize(self, fmap, stride=2):
self.bn0 = L.batch_norm()
self.activ = L.activation(M.PARAM_RELU)
self.c1 = L.conv2D(3, fmap, stride=stride, pad='VALID', usebias=False)
self.bn1 = L.batch_norm()
self.c2 = L.conv2D(3, fmap, pad='VALID', usebias=False)
# shortcut conv
self.c3 = L.conv2D(1, fmap, stride=stride, usebias=False)
def initialize(self, outsize, usebias=True, batch_norm=False, activation=-1): self.fclayer = L.fcLayer(outsize,usebias=usebias) self.batch_norm = batch_norm self.activation_ = activation if batch_norm: self.bn = L.batch_norm() if activation!=-1: self.activation = L.activation(activation)
def initialize(self, size, outchn, activation=-1, stride=1, usebias=True, pad='SAME', batch_norm=False): self.deconv = L.deconv2D(size,outchn,stride=stride,usebias=usebias,pad=pad, name=None) self.batch_norm = batch_norm self.activation_ = activation if batch_norm: self.bn = L.batch_norm() if activation!=-1: self.activation = L.activation(activation)
def initialize(self, outsize, adj_mtx=None, adj_fn=None, usebias=True, activation=-1, batch_norm=False): self.GCL = L.graphConvLayer(outsize, adj_mtx=adj_mtx, adj_fn=adj_fn, usebias=usebias) self.batch_norm = batch_norm self.activation_ = activation if batch_norm: self.bn = L.batch_norm() if activation!=-1: self.activation = L.activation(activation)
def initialize(self):
self.bn0 = L.batch_norm()
self.activ = L.activation(M.PARAM_RELU)
self.c1 = L.conv2D(3, 512, dilation_rate=2, pad='VALID', usebias=False)
self.bn1 = L.batch_norm()
self.c2 = L.conv2D(3,
1024,
dilation_rate=2,
pad='VALID',
usebias=False)
def initialize(self, outchn, stride=1, ratio=4, activation=PARAM_RELU): self.outchn = outchn # self.stride = stride self.activ = L.activation(activation) self.bn = L.batch_norm() self.l1 = ConvLayer(1, outchn//ratio, activation=PARAM_RELU, batch_norm=True) self.l2 = ConvLayer(3, outchn//ratio, activation=PARAM_RELU, batch_norm=True, stride=stride) self.l3 = ConvLayer(1, outchn) self.shortcut_conv = ConvLayer(1, outchn, activation=PARAM_RELU, stride=stride) self.shortcut_pool = L.maxpoolLayer(stride)
def initialize(self, fmaps, dilation_rate):
self.dilation_rate = dilation_rate
self.bn0 = L.batch_norm()
self.activ = L.activation(M.PARAM_RELU)
self.c1 = L.conv2D(1, fmaps[0], pad='VALID', usebias=False)
self.bn1 = L.batch_norm()
self.c2 = L.conv2D(3,
fmaps[1],
pad='VALID',
usebias=False,
dilation_rate=dilation_rate)
self.bn2 = L.batch_norm()
self.c3 = L.conv2D(1, fmaps[2], pad='VALID', usebias=False)
# shortcut
self.c4 = L.conv2D(1, fmaps[2], pad='VALID', usebias=False)
def initialize(self, fmap):
self.bn0 = L.batch_norm()
self.activ = L.activation(M.PARAM_RELU)
self.c1 = L.conv2D(3, fmap, pad='VALID', usebias=False)
self.bn1 = L.batch_norm()
self.c2 = L.conv2D(3, fmap, pad='VALID', usebias=False)
