def __init__(self, args, inplanes, planes, stride=1, downsample=None):
super(Bottleneck, self).__init__()
self.conv1 = conv1x1(inplanes, planes)
self.bn1 = nn.BatchNorm2d(planes)
self.relu1 = ReLU(args, inplace=True) # ReLuPCA(args, planes)
self.conv2 = conv3x3(planes, planes, stride)
self.bn2 = nn.BatchNorm2d(planes)
self.relu2 = ReLU(args, inplace=True) # ReLuPCA(args, planes)
self.conv3 = conv1x1(planes, planes * self.expansion)
self.bn3 = nn.BatchNorm2d(planes * self.expansion)
self.relu3 = ReLU(args, inplace=True) # ReLuPCA(args, planes)
self.downsample = downsample
self.stride = stride
def __init__(self, args, in_planes, planes, stride=1, downsample=None):
super(BasicBlock, self).__init__()
self.conv1 = conv3x3(args, in_planes, planes, stride)
self.bn1 = BatchNorm2d(planes)
self.relu1 = ReLU(args, inplace=True) # ReLuPCA(args, planes)
self.conv2 = conv3x3(args, planes, planes)
self.bn2 = BatchNorm2d(planes)
self.downsample = downsample
self.relu2 = ReLU(args, inplace=True) # ReLuPCA(args, planes)
self.stride = stride
def __init__(self, depth, args):
super(ResNetCifar, self).__init__()
num_classes = args.nClasses
assert (depth - 2) % 6 == 0, 'Depth should be 6n + 2'
n = (depth - 2) // 6
self.name = args.model
self.depth = depth - 1
self.dataset = args.dataset
block = BasicBlock
self.inplanes = 64
fmaps = [64, 128, 256] # CIFAR10
self.conv = nn.Conv2d(3, 64, kernel_size=3, stride=1, padding=1, bias=False)
self.bn = nn.BatchNorm2d(64)
self.relu = ReLU(args, inplace=True) # ReLuPCA(args, 64)
self.layer1 = self._make_layer(args, block, fmaps[0], n, stride=1)
self.layer2 = self._make_layer(args, block, fmaps[1], n, stride=2)
self.layer3 = self._make_layer(args, block, fmaps[2], n, stride=2)
self.avgpool = nn.AvgPool2d(kernel_size=8, stride=1)
self.flatten = flatten
self.fc = nn.Linear(fmaps[2] * 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')
# nn.init.kaiming_uniform_(m.weight, mode='fan_out', nonlinearity='relu')
elif isinstance(m, nn.BatchNorm2d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
def __init__(self, args, in_planes, planes, stride=1, downsample=None):
super(BasicBlock, self).__init__()
self.conv1 = conv3x3(in_planes, planes, stride)
self.bn1 = nn.BatchNorm2d(planes)
self.relu1 = ReLU(args, inplace=True, channel_count=planes,
entropy_approximation=args.ea) # ReLuPCA(args, planes)
self.conv2 = conv3x3(planes, planes)
self.bn2 = nn.BatchNorm2d(planes)
self.relu2 = ReLU(args, inplace=True, channel_count=planes,
entropy_approximation=args.ea) # ReLuPCA(args, planes)
self.downsample = downsample
self.stride = stride
def __init__(self, args, inp, oup, stride, expand_ratio):
super(InvertedResidual, self).__init__()
self.stride = stride
assert stride in [1, 2]
hidden_dim = int(round(inp * expand_ratio))
self.use_res_connect = self.stride == 1 and inp == oup
if expand_ratio == 1:
self.conv = nn.Sequential(
# dw
Conv2d(args,
hidden_dim,
hidden_dim,
3,
stride,
1,
groups=hidden_dim,
bias=False),
BatchNorm2d(hidden_dim),
ReLU(args, inplace=True, relu6=True),
# pw-linear
Conv2d(args, hidden_dim, oup, 1, 1, 0, bias=False),
BatchNorm2d(oup),
)
else:
self.conv = nn.Sequential(
# pw
Conv2d(args, inp, hidden_dim, 1, 1, 0, bias=False),
BatchNorm2d(hidden_dim),
ReLU(args, inplace=True, relu6=True),
# dw
Conv2d(args,
hidden_dim,
hidden_dim,
3,
stride,
1,
groups=hidden_dim,
bias=False),
BatchNorm2d(hidden_dim),
ReLU(args, inplace=True, relu6=True),
# pw-linear
Conv2d(args, hidden_dim, oup, 1, 1, 0, bias=False),
BatchNorm2d(oup),
)
def Layer(self, input_):
"""
This function creates the components inside a composite layer
of a Dense Block.
"""
with tf.variable_scope("Composite"):
next_layer = BatchNorm(input_, isTrain=self.isTrain)
next_layer = ReLU(next_layer)
next_layer = Conv(next_layer,
kernel_size=3,
stride=1,
output_channels=self.growth_rate)
next_layer = DropOut(next_layer, isTrain=self.isTrain, rate=0.2)
return next_layer
def __init__(self, block, layers, args, zero_init_residual=False):
super(ResNetImagenet, self).__init__()
num_classes = 1000
self.name = args.model
self.inplanes = 64
self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,
bias=False)
self.bn1 = nn.BatchNorm2d(64)
self.relu = ReLU(args, inplace=True, channel_count=64, entropy_approximation=args.ea) # ReLuPCA(args, planes)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
self.layer1 = self._make_layer(args, block, 64, layers[0])
self.layer2 = self._make_layer(args, block, 128, layers[1], stride=2)
self.layer3 = self._make_layer(args, block, 256, layers[2], stride=2)
self.layer4 = self._make_layer(args, block, 512, layers[3], stride=2)
self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
self.fc = nn.Linear(512 * block.expansion, num_classes)
def __init__(self, block, layers, args, depth, zero_init_residual=False):
super(ResNetImagenet, self).__init__()
num_classes = 1000
self.name = args.model
if args.noQuantEdges:
self.depth = depth - 2
else:
self.depth = depth - 1
self.inplanes = 64
self.conv1 = Conv2d(args, 3, 64, kernel_size=7, stride=2, padding=3,
bias=False)
self.bn1 = BatchNorm2d(64)
self.relu = ReLU(args, inplace=True) # ReLuPCA(args, planes)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
self.layer1 = self._make_layer(args, block, 64, layers[0])
self.layer2 = self._make_layer(args, block, 128, layers[1], stride=2)
self.layer3 = self._make_layer(args, block, 256, layers[2], stride=2)
self.layer4 = self._make_layer(args, block, 512, layers[3], stride=2)
self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
self.fc = nn.Linear(512 * block.expansion, num_classes)
self.quantBlocks = self.buildQuantBlocks(args.noQuantEdges)
def conv_1x1_bn(inp, oup, args):
return nn.Sequential(Conv2d(args, inp, oup, 1, 1, 0, bias=False),
BatchNorm2d(oup), ReLU(args, inplace=True,
relu6=True))