def __init__(self,
codeword_multiplier,
sparsity_multiplier,
centers,
intercept,
block,
nblocks,
growth_rate=12,
reduction=0.5,
nb_class=10):
super(GlobalLocalLabelDenseNet, self).__init__()
in_dims = []
self.growth_rate = growth_rate
num_planes = 2 * growth_rate
self.conv1 = nn.Conv2d(3, num_planes, kernel_size=3, padding=1)
self.dense1 = self._make_dense_layers(block, num_planes, nblocks[0])
num_planes += nblocks[0] * growth_rate
out_planes = int(math.floor(num_planes * reduction))
self.trans1 = Transition(num_planes, out_planes)
num_planes = out_planes
in_dims.append(num_planes)
self.dense2 = self._make_dense_layers(block, num_planes, nblocks[1])
num_planes += nblocks[1] * growth_rate
out_planes = int(math.floor(num_planes * reduction))
self.trans2 = Transition(num_planes, out_planes)
num_planes = out_planes
in_dims.append(num_planes)
self.dense3 = self._make_dense_layers(block, num_planes, nblocks[2])
num_planes += nblocks[2] * growth_rate
out_planes = int(math.floor(num_planes * reduction))
self.trans3 = Transition(num_planes, out_planes)
num_planes = out_planes
self.dense4 = self._make_dense_layers(block, num_planes, nblocks[3])
num_planes += nblocks[3] * growth_rate
in_dims.append(num_planes)
self.bn = nn.BatchNorm2d(num_planes)
self.linear = nn.Linear(num_planes, nb_class)
if centers is None:
centers = [
None,
] * 3
if intercept is None:
intercept = [
None,
] * 3
self.sparse_layers = nn.ModuleList()
sparsity_multiplier *= codeword_multiplier
self.sparse_layers.append(
Layers.GlobalLocalLabel(in_dims[0],
in_dims[0] * codeword_multiplier,
int(in_dims[0] * sparsity_multiplier),
centers=centers[0],
intercept=intercept[0]))
self.sparse_layers.append(
Layers.GlobalLocalLabel(in_dims[1],
in_dims[1] * codeword_multiplier,
int(in_dims[1] * sparsity_multiplier),
centers=centers[1],
intercept=intercept[1]))
self.sparse_layers.append(
Layers.GlobalLocalLabel(in_dims[2],
in_dims[2] * codeword_multiplier,
int(in_dims[2] * sparsity_multiplier),
centers=centers[2],
intercept=intercept[2]))
self.initialize()
def __init__(self,
codeword_multiplier,
sparsity_multiplier,
block,
layers,
num_classes=1000,
zero_init_residual=False,
groups=1,
width_per_group=64,
replace_stride_with_dilation=None,
norm_layer=None):
super(GlobalLocalLabelResNet, self).__init__()
if norm_layer is None:
norm_layer = nn.BatchNorm2d
self._norm_layer = norm_layer
in_dims = []
self.inplanes = 64
self.dilation = 1
if replace_stride_with_dilation is None:
# each element in the tuple indicates if we should replace
# the 2x2 stride with a dilated convolution instead
replace_stride_with_dilation = [False, False, False]
if len(replace_stride_with_dilation) != 3:
raise ValueError("replace_stride_with_dilation should be None "
"or a 3-element tuple, got {}".format(
replace_stride_with_dilation))
self.groups = groups
self.base_width = width_per_group
self.conv1 = nn.Conv2d(3,
self.inplanes,
kernel_size=7,
stride=2,
padding=3,
bias=False)
self.bn1 = norm_layer(self.inplanes)
self.relu = nn.ReLU(inplace=True)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
in_dims.append(self.inplanes)
self.layer1 = self._make_layer(block, 64, layers[0])
self.layer2 = self._make_layer(block,
128,
layers[1],
stride=2,
dilate=replace_stride_with_dilation[0])
in_dims.append(self.inplanes)
self.layer3 = self._make_layer(block,
256,
layers[2],
stride=2,
dilate=replace_stride_with_dilation[1])
in_dims.append(self.inplanes)
self.layer4 = self._make_layer(block,
512,
layers[3],
stride=2,
dilate=replace_stride_with_dilation[2])
in_dims.append(self.inplanes)
in_dims.append(self.inplanes)
self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
self.fc = nn.Linear(512 * block.expansion, num_classes)
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight,
mode='fan_out',
nonlinearity='relu')
elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
# Zero-initialize the last BN in each residual branch,
# so that the residual branch starts with zeros, and each residual block behaves like an identity.
# This improves the model by 0.2~0.3% according to https://arxiv.org/abs/1706.02677
if zero_init_residual:
for m in self.modules():
if isinstance(m, Bottleneck):
nn.init.constant_(m.bn3.weight, 0)
elif isinstance(m, BasicBlock):
nn.init.constant_(m.bn2.weight, 0)
sparsity_multiplier *= codeword_multiplier
self.sparse_layers = nn.ModuleList()
for input_dim in in_dims:
self.sparse_layers.append(
Layers.GlobalLocalLabel(input_dim,
input_dim * codeword_multiplier,
input_dim * sparsity_multiplier))
self.in_dims = in_dims
