def convolution(x):
x = self.convolution[0](x)
x = self.convolution[1](x)
x = self.convolution[2](x)
x = M.max_pooling_2d(x, 3, 2)
x = self.convolution[3](x)
x = self.convolution[4](x)
x = M.max_pooling_2d(x, 3, 2)
return x
def __init__(self, pretrained_model='auto'):
super(GoogLeNet, self).__init__(
conv1=Convolution2D(3, 64, 7, stride=2, pad=3),
conv2_reduce=Convolution2D(64, 64, 1),
conv2=Convolution2D(64, 192, 3, stride=1, pad=1),
inc3a=Inception(192, 64, 96, 128, 16, 32, 32),
inc3b=Inception(256, 128, 128, 192, 32, 96, 64),
inc4a=Inception(480, 192, 96, 208, 16, 48, 64),
inc4b=Inception(512, 160, 112, 224, 24, 64, 64),
inc4c=Inception(512, 128, 128, 256, 24, 64, 64),
inc4d=Inception(512, 112, 144, 288, 32, 64, 64),
inc4e=Inception(528, 256, 160, 320, 32, 128, 128),
inc5a=Inception(832, 256, 160, 320, 32, 128, 128),
inc5b=Inception(832, 384, 192, 384, 48, 128, 128),
loss3_fc=Linear(1024, 1000),
loss1_conv=Convolution2D(512, 128, 1),
loss1_fc1=Linear(4 * 4 * 128, 1024),
loss1_fc2=Linear(1024, 1000),
loss2_conv=Convolution2D(528, 128, 1),
loss2_fc1=Linear(4 * 4 * 128, 1024),
loss2_fc2=Linear(1024, 1000),
)
if pretrained_model == 'auto':
_retrieve(
'bvlc_googlenet.npz',
'http://dl.caffe.berkeleyvision.org/bvlc_googlenet.caffemodel',
self)
elif pretrained_model:
npz.load_npz(pretrained_model, self)
self.functions = OrderedDict([
('conv1', [self.conv1, relu]),
('pool1', [
lambda x: max_pooling_2d(x, ksize=3, stride=2),
lambda x: local_response_normalization(x, n=5)
]), ('conv2_reduce', [self.conv2_reduce, relu]),
('conv2', [self.conv2, relu]),
('pool2', [
lambda x: local_response_normalization(x, n=5),
lambda x: max_pooling_2d(x, ksize=3, stride=2)
]), ('inc3a', [self.inc3a]), ('inc3b', [self.inc3b]),
('pool3', [lambda x: max_pooling_2d(x, ksize=3, stride=2)]),
('inc4a', [self.inc4a]), ('inc4b', [self.inc4b]),
('inc4c', [self.inc4c]), ('inc4d', [self.inc4d]),
('inc4e', [self.inc4e]),
('pool4', [lambda x: max_pooling_2d(x, ksize=3, stride=2)]),
('inc5a', [self.inc5a]), ('inc5b', [self.inc5b]),
('pool6', [lambda x: average_pooling_2d(x, ksize=7, stride=1)]),
('prob', [lambda x: dropout(x, ratio=0.4), self.loss3_fc])
])
def __call__(self, x):
h = max_pooling_2d.max_pooling_2d(relu.relu(self.conv1(x)),
3,
stride=2)
h = max_pooling_2d.max_pooling_2d(relu.relu(self.conv2(h)),
3,
stride=2)
h = relu.relu(self.conv3(h))
h = relu.relu(self.conv4(h))
h = max_pooling_2d.max_pooling_2d(relu.relu(self.conv5(h)),
3,
stride=2)
h = self.fc6(h)
h = self.fc7(h)
return self.fc8(h)
def __call__(self, x):
outs = []
if self.out1 > 0:
h1 = self.conv1(x)
h1 = self.conv1n(h1)
h1 = relu.relu(h1)
outs.append(h1)
h3 = relu.relu(self.proj3n(self.proj3(x)))
h3 = relu.relu(self.conv3n(self.conv3(h3)))
outs.append(h3)
h33 = relu.relu(self.proj33n(self.proj33(x)))
h33 = relu.relu(self.conv33an(self.conv33a(h33)))
h33 = relu.relu(self.conv33bn(self.conv33b(h33)))
outs.append(h33)
if self.pooltype == 'max':
p = max_pooling_2d.max_pooling_2d(x, 3, stride=self.stride, pad=1,
cover_all=False)
else:
p = average_pooling_2d.average_pooling_2d(x, 3, stride=self.stride,
pad=1)
if self.proj_pool is not None:
p = relu.relu(self.poolpn(self.poolp(p)))
outs.append(p)
y = concat.concat(outs, axis=1)
return y
def __call__(self, x):
test = not self.train
outs = []
if self.out1 > 0:
h1 = self.conv1(x)
h1 = self.conv1n(h1, test=test)
h1 = relu.relu(h1)
outs.append(h1)
h3 = relu.relu(self.proj3n(self.proj3(x), test=test))
h3 = relu.relu(self.conv3n(self.conv3(h3), test=test))
outs.append(h3)
h33 = relu.relu(self.proj33n(self.proj33(x), test=test))
h33 = relu.relu(self.conv33an(self.conv33a(h33), test=test))
h33 = relu.relu(self.conv33bn(self.conv33b(h33), test=test))
outs.append(h33)
if self.pooltype == 'max':
p = max_pooling_2d.max_pooling_2d(x, 3, stride=self.stride, pad=1)
else:
p = average_pooling_2d.average_pooling_2d(x, 3, stride=self.stride,
pad=1)
if self.proj_pool is not None:
p = relu.relu(self.poolpn(self.poolp(p), test=test))
outs.append(p)
y = concat.concat(outs, axis=1)
return y
def __init__(self, pretrained_model='auto'):
if pretrained_model:
# As a sampling process is time-consuming,
# we employ a zero initializer for faster computation.
kwargs = {'initialW': constant.Zero()}
else:
# employ default initializers used in the original paper
kwargs = {'initialW': normal.HeNormal(scale=1.0)}
super(ResNet50Layers, self).__init__(
conv1=Convolution2D(3, 64, 7, 2, 3, **kwargs),
bn1=BatchNormalization(64),
res2=BuildingBlock(3, 64, 64, 256, 1, **kwargs),
res3=BuildingBlock(4, 256, 128, 512, 2, **kwargs),
res4=BuildingBlock(6, 512, 256, 1024, 2, **kwargs),
res5=BuildingBlock(3, 1024, 512, 2048, 2, **kwargs),
fc6=Linear(2048, 1000),
)
if pretrained_model == 'auto':
_retrieve(
'ResNet-50-model.npz', 'ResNet-50-model.caffemodel', self)
elif pretrained_model:
npz.load_npz(pretrained_model, self)
self.functions = OrderedDict([
('conv1', [self.conv1, self.bn1, relu]),
('pool1', [lambda x: max_pooling_2d(x, ksize=3, stride=2)]),
('res2', [self.res2]),
('res3', [self.res3]),
('res4', [self.res4]),
('res5', [self.res5]),
('pool5', [_global_average_pooling_2d]),
('fc6', [self.fc6]),
('prob', [softmax]),
])
def functions(self):
return collections.OrderedDict([
('conv1', [self.conv1, self.bn1, relu]),
('pool1', [lambda x: max_pooling_2d(x, ksize=3, stride=2)]),
('res2', [self.res2]), ('res3', [self.res3]),
('res4', [self.res4]), ('res5', [self.res5])
])
def forward(self, x):
outs = []
if self.out1 > 0:
h1 = self.conv1(x)
h1 = self.conv1n(h1)
h1 = relu.relu(h1)
outs.append(h1)
h3 = relu.relu(self.proj3n(self.proj3(x)))
h3 = relu.relu(self.conv3n(self.conv3(h3)))
outs.append(h3)
h33 = relu.relu(self.proj33n(self.proj33(x)))
h33 = relu.relu(self.conv33an(self.conv33a(h33)))
h33 = relu.relu(self.conv33bn(self.conv33b(h33)))
outs.append(h33)
if self.pooltype == 'max':
p = max_pooling_2d.max_pooling_2d(x, 3, stride=self.stride, pad=1,
cover_all=False)
else:
p = average_pooling_2d.average_pooling_2d(x, 3, stride=self.stride,
pad=1)
if self.proj_pool is not None:
p = relu.relu(self.poolpn(self.poolp(p)))
outs.append(p)
y = concat.concat(outs, axis=1)
return y
def __call__(self, x):
out1 = self.f.conv1(x)
out3 = self.f.conv3(relu.relu(self.f.proj3(x)))
out5 = self.f.conv5(relu.relu(self.f.proj5(x)))
pool = self.f.projp(
max_pooling_2d.max_pooling_2d(x, 3, stride=1, pad=1))
y = relu.relu(concat.concat((out1, out3, out5, pool), axis=1))
return y
def __call__(self, x):
out1 = self.f.conv1(x)
out3 = self.f.conv3(relu.relu(self.f.proj3(x)))
out5 = self.f.conv5(relu.relu(self.f.proj5(x)))
pool = self.f.projp(max_pooling_2d.max_pooling_2d(
x, 3, stride=1, pad=1))
y = relu.relu(concat.concat((out1, out3, out5, pool), axis=1))
return y
def __call__(self, x):
h = relu.relu(self.conv1(x), self.use_cudnn)
h = max_pooling_2d.max_pooling_2d(h, 2, stride=2, use_cudnn=self.use_cudnn)
h = relu.relu(self.conv2(h), self.use_cudnn)
h = max_pooling_2d.max_pooling_2d(h, 2, stride=2, use_cudnn=self.use_cudnn)
h = relu.relu(self.conv3_1(h), self.use_cudnn)
h = relu.relu(self.conv3_2(h), self.use_cudnn)
h = max_pooling_2d.max_pooling_2d(h, 2, stride=2, use_cudnn=self.use_cudnn)
h = relu.relu(self.conv4_1(h), self.use_cudnn)
h = relu.relu(self.conv4_2(h), self.use_cudnn)
h = max_pooling_2d.max_pooling_2d(h, 2, stride=2, use_cudnn=self.use_cudnn)
h = relu.relu(self.conv5_1(h), self.use_cudnn)
h = relu.relu(self.conv5_2(h), self.use_cudnn)
h = max_pooling_2d.max_pooling_2d(h, 2, stride=2, use_cudnn=self.use_cudnn)
h = self.fc6(h)
h = self.fc7(h)
return self.fc8(h)
def functions(self):
return collections.OrderedDict([
('conv1', [self.conv1, relu]),
('pool1', [lambda x: max_pooling_2d(x, 3, stride=2)]),
('fire2', [self.fire2]),
('fire3', [self.fire3]),
('pool2', [lambda x: max_pooling_2d(x, 3, stride=2)]),
('fire4', [self.fire4]),
('fire5', [self.fire5]),
('pool3', [lambda x: max_pooling_2d(x, 3, stride=2)]),
('fire6', [self.fire6]),
('fire7', [self.fire7]),
('fire8', [self.fire8]),
('fire9', [self.fire9, dropout]),
('conv10', [self.conv10, relu]),
('pool4', [lambda x: average_pooling_2d(x, 13)]),
('prob', [lambda x: reshape(x, (-1, 1000))]),
])
def functions(self):
return collections.OrderedDict([
('conv1', [self.conv1, self.bn1, relu]),
('pool1', [lambda x: max_pooling_2d(x, ksize=3, stride=2)]),
('res2', [self.res2]),
('res3', [self.res3]),
('res4', [self.res4]),
('res5', [self.res5]),
('pool5', [_global_average_pooling_2d]),
('fc6', [self.fc6]),
('prob', [softmax]),
])
def functions(self):
return collections.OrderedDict([
('conv1', [self.conv1, self.bn1, relu]),
('pool1', [lambda x: max_pooling_2d(x, ksize=3, stride=2)]),
('res2', [self.res2]),
('res3', [self.res3]),
('res4', [self.res4]),
('res5', [self.res5]),
('pool5', [_global_average_pooling_2d]),
('fc6', [self.fc6]),
('prob', [softmax]),
])
def __call__(self, x):
h0 = max_pooling_2d.max_pooling_2d(relu.relu(self.conv1a(x)),
3,
stride=2)
h1 = max_pooling_2d.max_pooling_2d(relu.relu(self.conv1b(x)),
3,
stride=2)
h0 = max_pooling_2d.max_pooling_2d(relu.relu(self.conv2a(h0)),
3,
stride=2)
h1 = max_pooling_2d.max_pooling_2d(relu.relu(self.conv2b(h1)),
3,
stride=2)
h2 = relu.relu(self.conv3a(h0) + self.conv3b(h1))
h3 = relu.relu(self.conv3c(h0) + self.conv3d(h1))
h2 = relu.relu(self.conv4a(h2))
h3 = relu.relu(self.conv4b(h3))
h2 = max_pooling_2d.max_pooling_2d(relu.relu(self.conv5a(h2)),
3,
stride=2)
h3 = max_pooling_2d.max_pooling_2d(relu.relu(self.conv5b(h3)),
3,
stride=2)
h3 = self.fc6a(h2) + self.fc6b(h3)
h3 = self.fc7(h3)
return self.fc8(h3)
def functions(self):
return collections.OrderedDict([
('conv1', [self.conv1, self.bn1, relu]),
('pool1', [lambda x: max_pooling_2d(x, ksize=3, stride=2)]),
('res2', [self.res2]),
('res3', [self.res3]),
('res4', [self.res4]),
('res5', [self.res5]),
#('pool5', [_global_average_pooling_2d]), #not for us...
#('fc6', [self.fc6]),
('score_fr', [self.score_fr]),
('upscore', [self.upscore]),
#('prob', [softmax]), #the other fcn doesn't have this, as there its in the sofmax crossentropy
])
def forward(self, x):
"""Computes the output of the Inception module.
Args:
x (~chainer.Variable): Input variable.
Returns:
Variable: Output variable. Its array has the same spatial size and
the same minibatch size as the input array. The channel dimension
has size ``out1 + out3 + out5 + proj_pool``.
"""
out1 = self.conv1(x)
out3 = self.conv3(relu.relu(self.proj3(x)))
out5 = self.conv5(relu.relu(self.proj5(x)))
pool = self.projp(max_pooling_2d.max_pooling_2d(x, 3, stride=1, pad=1))
y = relu.relu(concat.concat((out1, out3, out5, pool), axis=1))
return y
def __call__(self, x):
h = self.bn1(self.conv1(x), test=not self.train)
h = max_pooling_2d.max_pooling_2d(relu.relu(h),
3,
stride=2,
use_cudnn=self.use_cudnn)
h = self.res2(h, self.train)
h = self.res3(h, self.train)
h = self.res4(h, self.train)
h = self.res5(h, self.train)
h = average_pooling_2d.average_pooling_2d(h,
7,
stride=1,
use_cudnn=self.use_cudnn)
h = self.fc(h)
return h
def __call__(self, x, test=None):
"""Computes the output of the InceptionBN module.
Args:
x (Variable): An input variable.
test (bool): If ``True``, batch normalization layers run in testing
mode; if ``test`` is omitted, ``not self.train`` is used as
``test``.
"""
if test is None:
test = not self.train
outs = []
if self.out1 > 0:
h1 = self.conv1(x)
h1 = self.conv1n(h1, test=test)
h1 = relu.relu(h1)
outs.append(h1)
h3 = relu.relu(self.proj3n(self.proj3(x), test=test))
h3 = relu.relu(self.conv3n(self.conv3(h3), test=test))
outs.append(h3)
h33 = relu.relu(self.proj33n(self.proj33(x), test=test))
h33 = relu.relu(self.conv33an(self.conv33a(h33), test=test))
h33 = relu.relu(self.conv33bn(self.conv33b(h33), test=test))
outs.append(h33)
if self.pooltype == 'max':
p = max_pooling_2d.max_pooling_2d(x,
3,
stride=self.stride,
pad=1,
cover_all=False)
else:
p = average_pooling_2d.average_pooling_2d(x,
3,
stride=self.stride,
pad=1)
if self.proj_pool is not None:
p = relu.relu(self.poolpn(self.poolp(p), test=test))
outs.append(p)
y = concat.concat(outs, axis=1)
return y
def __init__(self, pretrained_model, n_layers):
super(ResNetLayers, self).__init__()
if pretrained_model:
# As a sampling process is time-consuming,
# we employ a zero initializer for faster computation.
kwargs = {'initialW': constant.Zero()}
else:
# employ default initializers used in the original paper
kwargs = {'initialW': normal.HeNormal(scale=1.0)}
if n_layers == 50:
block = [3, 4, 6, 3]
elif n_layers == 101:
block = [3, 4, 23, 3]
elif n_layers == 152:
block = [3, 8, 36, 3]
else:
raise ValueError('The n_layers argument should be either 50, 101,'
' or 152, but {} was given.'.format(n_layers))
with self.init_scope():
self.conv1 = Convolution2D(3, 64, 7, 2, 3, **kwargs)
self.bn1 = BatchNormalization(64)
self.res2 = BuildingBlock(block[0], 64, 64, 256, 1, **kwargs)
self.res3 = BuildingBlock(block[1], 256, 128, 512, 2, **kwargs)
self.res4 = BuildingBlock(block[2], 512, 256, 1024, 2, **kwargs)
self.res5 = BuildingBlock(block[3], 1024, 512, 2048, 2, **kwargs)
self.fc6 = Linear(2048, 1000)
if pretrained_model and pretrained_model.endswith('.caffemodel'):
_retrieve(n_layers, 'ResNet-{}-model.npz'.format(n_layers),
pretrained_model, self)
elif pretrained_model:
npz.load_npz(pretrained_model, self)
self.functions = collections.OrderedDict([
('conv1', [self.conv1, self.bn1, relu]),
('pool1', [lambda x: max_pooling_2d(x, ksize=3, stride=2)]),
('res2', [self.res2]),
('res3', [self.res3]),
('res4', [self.res4]),
('res5', [self.res5]),
('pool5', [_global_average_pooling_2d]),
('fc6', [self.fc6]),
('prob', [softmax]),
])
def __init__(self, pretrained_model, n_layers):
if pretrained_model:
# As a sampling process is time-consuming,
# we employ a zero initializer for faster computation.
kwargs = {'initialW': constant.Zero()}
else:
# employ default initializers used in the original paper
kwargs = {'initialW': normal.HeNormal(scale=1.0)}
if n_layers == 50:
block = [3, 4, 6, 3]
elif n_layers == 101:
block = [3, 4, 23, 3]
elif n_layers == 152:
block = [3, 8, 36, 3]
else:
raise ValueError('The n_layers argument should be either 50, 101,'
' or 152, but {} was given.'.format(n_layers))
super(ResNetLayers, self).__init__(
conv1=Convolution2D(3, 64, 7, 2, 3, **kwargs),
bn1=BatchNormalization(64),
res2=BuildingBlock(block[0], 64, 64, 256, 1, **kwargs),
res3=BuildingBlock(block[1], 256, 128, 512, 2, **kwargs),
res4=BuildingBlock(block[2], 512, 256, 1024, 2, **kwargs),
res5=BuildingBlock(block[3], 1024, 512, 2048, 2, **kwargs),
fc6=Linear(2048, 1000),
)
if pretrained_model and pretrained_model.endswith('.caffemodel'):
_retrieve(n_layers, 'ResNet-{}-model.npz'.format(n_layers),
pretrained_model, self)
elif pretrained_model:
npz.load_npz(pretrained_model, self)
self.functions = collections.OrderedDict([
('conv1', [self.conv1, self.bn1, relu]),
('pool1', [lambda x: max_pooling_2d(x, ksize=3, stride=2)]),
('res2', [self.res2]),
('res3', [self.res3]),
('res4', [self.res4]),
('res5', [self.res5]),
('pool5', [_global_average_pooling_2d]),
('fc6', [self.fc6]),
('prob', [softmax]),
])
def __call__(self, x):
"""Computes the output of the Inception module.
Args:
x (~chainer.Variable): Input variable.
Returns:
Variable: Output variable. Its array has the same spatial size and
the same minibatch size as the input array. The channel dimension
has size ``out1 + out3 + out5 + proj_pool``.
"""
out1 = self.conv1(x)
out3 = self.conv3(relu.relu(self.proj3(x)))
out5 = self.conv5(relu.relu(self.proj5(x)))
pool = self.projp(max_pooling_2d.max_pooling_2d(
x, 3, stride=1, pad=1))
y = relu.relu(concat.concat((out1, out3, out5, pool), axis=1))
return y
def __call__(self, x, test=None):
"""Computes the output of the InceptionBN module.
Args:
x (Variable): An input variable.
test (bool): If ``True``, batch normalization layers run in testing
mode; if ``test`` is omitted, ``not self.train`` is used as
``test``.
"""
if test is None:
test = not self.train
outs = []
if self.out1 > 0:
h1 = self.conv1(x)
h1 = self.conv1n(h1, test=test)
h1 = relu.relu(h1)
outs.append(h1)
h3 = relu.relu(self.proj3n(self.proj3(x), test=test))
h3 = relu.relu(self.conv3n(self.conv3(h3), test=test))
outs.append(h3)
h33 = relu.relu(self.proj33n(self.proj33(x), test=test))
h33 = relu.relu(self.conv33an(self.conv33a(h33), test=test))
h33 = relu.relu(self.conv33bn(self.conv33b(h33), test=test))
outs.append(h33)
if self.pooltype == 'max':
p = max_pooling_2d.max_pooling_2d(x, 3, stride=self.stride, pad=1,
cover_all=False)
else:
p = average_pooling_2d.average_pooling_2d(x, 3, stride=self.stride,
pad=1)
if self.proj_pool is not None:
p = relu.relu(self.poolpn(self.poolp(p), test=test))
outs.append(p)
y = concat.concat(outs, axis=1)
return y
def _max_pooling_2d(x):
return max_pooling_2d(x, ksize=2)
def _max_pooling_2d(x):
return max_pooling_2d(x, ksize=2)