def __init__(self, num_classes=1000, version='v1'):
super(GoogLeNetV12, self).__init__()
self.conv1 = BasicConv2d(3, 64, kernel_size=7, stride=2, padding=3)
self.maxpool1 = nn.MaxPool2d(3, 2, ceil_mode=True)
self.conv2 = BasicConv2d(64, 64, kernel_size=1)
self.conv3 = BasicConv2d(64, 192, kernel_size=3, stride=1, padding=1)
self.maxpool2 = nn.MaxPool2d(3, 2, ceil_mode=True)
self.inception3a = Inception(192, 64, 96, 128, 16, 32, 32, version)
self.inception3b = Inception(256, 128, 128, 192, 32, 96, 64, version)
self.maxpool3 = nn.MaxPool2d(3, 2, ceil_mode=True)
self.inception4a = Inception(480, 192, 96, 208, 16, 48, 64, version)
self.inception4b = Inception(512, 160, 112, 224, 24, 64, 64, version)
self.inception4c = Inception(512, 128, 128, 256, 24, 64, 64, version)
self.inception4d = Inception(512, 112, 144, 288, 32, 64, 64, version)
self.inception4e = Inception(528, 256, 160, 320, 32, 128, 128, version)
self.maxpool4 = nn.MaxPool2d(3, 2, ceil_mode=True)
self.inception5a = Inception(832, 256, 160, 320, 32, 128, 128, version)
self.inception5b = Inception(832, 384, 192, 384, 48, 128, 128, version)
self.avgpool = nn.AvgPool2d(kernel_size=7, stride=1)
self.dropout = nn.Dropout(p=0.4)
self.fc = nn.Linear(1024, num_classes)
def __init__(self, in_channels, k, l, m, n):
super(Reduction_A, self).__init__()
self.branch2 = BasicConv2d(in_channels,
n,
kernel_size=3,
stride=2,
padding=0,
bias=False)
self.branch3 = nn.Sequential(
BasicConv2d(in_channels,
k,
kernel_size=1,
stride=1,
padding=0,
bias=False),
BasicConv2d(k, l, kernel_size=3, stride=1, padding=1, bias=False),
BasicConv2d(l, m, kernel_size=3, stride=2, padding=0, bias=False))
def __init__(self, in_channels, growth_rate, bn_size):
super(_DenseLayer, self).__init__()
self.add_module(
'norm_relu_conv1',
BasicConv2d(in_channels,
bn_size * growth_rate,
kernel_size=1,
stride=1,
bias=False))
self.add_module(
'norm_relu_conv2',
BasicConv2d(bn_size * growth_rate,
growth_rate,
kernel_size=3,
stride=1,
padding=1,
bias=False))
def __init__(self, in_channels, out_channels):
super(_Transition, self).__init__()
self.add_module(
'norm_relu_conv',
BasicConv2d(in_channels,
out_channels,
kernel_size=1,
stride=1,
bias=False))
self.add_module('pool', nn.AvgPool2d(kernel_size=2, stride=2))
def __init__(self, in_channels, b1, b2_n1, b2_n1x3, b2_n3x1, n1_linear):
super(InceptionC_res, self).__init__()
self.branch1 = BasicConv2d(in_channels,
b1,
kernel_size=1,
stride=1,
padding=0,
bias=False)
self.branch2 = nn.Sequential(
BasicConv2d(in_channels,
b2_n1,
kernel_size=1,
stride=1,
padding=0,
bias=False),
BasicConv2d(b2_n1,
b2_n1x3,
kernel_size=(1, 3),
stride=(1, 1),
padding=(0, 1),
bias=False),
BasicConv2d(b2_n1x3,
b2_n3x1,
kernel_size=(3, 1),
stride=(1, 1),
padding=(1, 0),
bias=False))
self.conv_linear = nn.Conv2d(b1 + b2_n3x1,
n1_linear,
kernel_size=1,
stride=1,
padding=0,
bias=False)
self.short_cut = nn.Sequential()
if in_channels != n1_linear:
self.short_cut = nn.Sequential(
nn.Conv2d(in_channels,
n1_linear,
kernel_size=1,
stride=1,
padding=0,
bias=False), nn.BatchNorm2d(n1_linear))
def __init__(self, num_classes = 1000, version = 'v3'):
# 299 x 299 x 3
super(GoogLeNetV3, self).__init__()
self.conv = nn.Sequential(
BasicConv2d(3, 32, kernel_size=3, stride = 2),
BasicConv2d(32, 32, kernel_size=3, stride = 1),
BasicConv2d(32, 64, kernel_size=3, stride=1, padding = 1),
nn.MaxPool2d(3,2),
BasicConv2d(64, 80, kernel_size=3, stride=1),
BasicConv2d(80, 192, kernel_size=3, stride=2),
BasicConv2d(192, 288, kernel_size=3, stride=1, padding = 1)
)
self.inception3a = Inceptionv3_modelA(288, 64, 64, 96, 48, 64, 64)
self.inception3b = Inceptionv3_modelA(288, 64, 64, 96, 48, 64, 64)
self.inception3c = Inceptionv3_modelA(288, 0, 192, 384, 64, 96, 288)
self.inception5a = Inceptionv3_modelB(768, 192, 160, 192, 160, 192, 192)
self.inception5b = Inceptionv3_modelB(768, 192, 160, 192, 160, 192, 192)
self.inception5c = Inceptionv3_modelB(768, 192, 160, 192, 160, 192, 192)
self.inception5d = Inceptionv3_modelB(768, 192, 160, 192, 160, 192, 192)
self.inception5e = Inceptionv3_modelB(768, 0, 192, 320, 192, 192, 768)
self.inception2a = Inceptionv3_modelC(1280, 320, 384, 384, 448, 384, 192)
self.inception2b = Inceptionv3_modelC(2048, 320, 384, 384, 448, 384, 192)
self.pool = nn.MaxPool2d(8,1)
self.fc = nn.Linear(2048, num_classes)
def __init__(self, in_channels, out1_channels, out3in_channels, out3_channels, out5in_channels, out5_channels, outpool_channels):
super(Inceptionv3_modelC, self).__init__()
self.branch1 = BasicConv2d(in_channels, out1_channels, kernel_size=1)
self.branch2_0 = BasicConv2d(in_channels, out3in_channels, kernel_size=1)
self.branch2_1 = BasicConv2d(out3in_channels, out3_channels, kernel_size=(3, 1), padding = (1,0))
self.branch2_2 = BasicConv2d(out3in_channels, out3_channels, kernel_size=(1, 3), padding = (0,1))
self.branch3_0 = BasicConv2d(in_channels, out5in_channels, kernel_size=1)
self.branch3_1 = nn.Sequential(
BasicConv2d(out5in_channels, out5_channels, kernel_size=(3, 3), padding=1),
BasicConv2d(out5_channels, out5_channels, kernel_size=(1, 3), padding = (0,1)),
)
self.branch3_2 = nn.Sequential(
BasicConv2d(out5in_channels, out5_channels, kernel_size=(3, 3), padding=1),
BasicConv2d(out5_channels, out5_channels, kernel_size=(3, 1), padding = (1,0)),
)
self.branch4 = nn.Sequential(
nn.MaxPool2d(3, stride=1, padding=1),
BasicConv2d(in_channels, outpool_channels, kernel_size=1)
)
def __init__(self,
in_channels,
out1_channels,
out3in_channels,
out3_channels,
out5in_channels,
out5_channels,
outpool_channels,
version='v1'):
super(Inception, self).__init__()
# 1x1 conv branch
self.branch1 = BasicConv2d(in_channels, out1_channels, kernel_size=1)
# 1x1 conv -> 3x3 conv branch
self.branch2 = nn.Sequential(
BasicConv2d(in_channels, out3in_channels, kernel_size=1),
BasicConv2d(out3in_channels,
out3_channels,
kernel_size=3,
padding=1))
# 1x1 conv -> 5x5 conv branch
# 1x1 conv -> 3x3 conv branch -> 3x3 conv branch
if version == 'v1':
self.branch3 = nn.Sequential(
BasicConv2d(in_channels, out5in_channels, kernel_size=1),
BasicConv2d(out5in_channels,
out5_channels,
kernel_size=3,
padding=1))
elif version == 'v2':
self.branch3 = nn.Sequential(
BasicConv2d(in_channels, out5in_channels, kernel_size=1),
BasicConv2d(out5in_channels,
out5_channels,
kernel_size=3,
padding=1),
BasicConv2d(out5_channels,
out5_channels,
kernel_size=3,
padding=1))
# 3x3 pool -> 1x1 conv branch
self.branch4 = nn.Sequential(
nn.MaxPool2d(3, stride=1, padding=1),
BasicConv2d(in_channels, outpool_channels, kernel_size=1))
def __init__(self, in_channels, b1, b2_n1, b2_n3, b3_n1, b3_n3_1, b3_n3_2,
n1_linear):
super(Inception_A_res, self).__init__()
self.branch1 = BasicConv2d(in_channels,
b1,
kernel_size=1,
stride=1,
padding=0,
bias=False)
self.branch2 = nn.Sequential(
BasicConv2d(in_channels,
b2_n1,
kernel_size=1,
stride=1,
padding=0,
bias=False),
BasicConv2d(b2_n1,
b2_n3,
kernel_size=3,
stride=1,
padding=1,
bias=False),
)
self.branch3 = nn.Sequential(
BasicConv2d(in_channels,
b3_n1,
kernel_size=1,
stride=1,
padding=0,
bias=False),
BasicConv2d(b3_n1,
b3_n3_1,
kernel_size=3,
stride=1,
padding=1,
bias=False),
BasicConv2d(b3_n3_1,
b3_n3_2,
kernel_size=3,
stride=1,
padding=1,
bias=False))
self.conv_linear = nn.Conv2d(b1 + b2_n3 + b3_n3_2,
n1_linear,
1,
1,
0,
bias=True)
self.short_cut = nn.Sequential()
if in_channels != n1_linear:
self.short_cut = nn.Sequential(
nn.Conv2d(in_channels,
n1_linear,
kernel_size=1,
stride=1,
padding=0,
bias=False), nn.BatchNorm2d(n1_linear))
def __init__(self, in_channels, b2_n1, b2_n3, b3_n1, b3_n3, b4_n1, b4_n3_1,
b4_n3_2):
super(Reduction_B_Res, self).__init__()
self.branch2 = nn.Sequential(
BasicConv2d(in_channels,
b2_n1,
kernel_size=1,
stride=1,
padding=0,
bias=False),
BasicConv2d(b2_n1,
b2_n3,
kernel_size=3,
stride=2,
padding=0,
bias=False),
)
self.branch3 = nn.Sequential(
BasicConv2d(in_channels,
b3_n1,
kernel_size=1,
stride=1,
padding=0,
bias=False),
BasicConv2d(b3_n1,
b3_n3,
kernel_size=3,
stride=2,
padding=0,
bias=False))
self.branch4 = nn.Sequential(
BasicConv2d(in_channels,
b4_n1,
kernel_size=1,
stride=1,
padding=0,
bias=False),
BasicConv2d(b4_n1,
b4_n3_1,
kernel_size=3,
stride=1,
padding=1,
bias=False),
BasicConv2d(b4_n3_1,
b4_n3_2,
kernel_size=3,
stride=2,
padding=0,
bias=False))
def __init__(self, in_channels, b1, b2, b3_n1, b3_n3, b4_n1, b4_n3):
super(Inception_A, self).__init__()
self.branch1 = nn.Sequential(
nn.AvgPool2d(3, 1, 1),
BasicConv2d(in_channels,
b1,
kernel_size=1,
stride=1,
padding=0,
bias=False))
self.branch2 = BasicConv2d(in_channels,
b2,
kernel_size=1,
stride=1,
padding=0,
bias=False)
self.branch3 = nn.Sequential(
BasicConv2d(in_channels,
b3_n1,
kernel_size=1,
stride=1,
padding=0,
bias=False),
BasicConv2d(b3_n1,
b3_n3,
kernel_size=3,
stride=1,
padding=1,
bias=False))
self.branch4 = nn.Sequential(
BasicConv2d(in_channels,
b4_n1,
kernel_size=1,
stride=1,
padding=0,
bias=False),
BasicConv2d(b4_n1,
b4_n3,
kernel_size=3,
stride=1,
padding=1,
bias=False),
BasicConv2d(b4_n3,
b4_n3,
kernel_size=3,
stride=1,
padding=1,
bias=False))
def __init__(self, in_channels, out1_channels, out3in_channels, out3_channels, out5in_channels, out5_channels, outpool_channels):
super(Inceptionv3_modelA, self).__init__()
self.out1_channels = out1_channels
stride = 1 if out1_channels > 0 else 2
padding = 1 if out1_channels > 0 else 0
if out1_channels > 0:
self.branch1 = BasicConv2d(in_channels, out1_channels, kernel_size=1)
# 1x1 conv -> 3x3 conv branch
self.branch2 = nn.Sequential(
BasicConv2d(in_channels, out3in_channels, kernel_size=1),
BasicConv2d(out3in_channels, out3_channels, kernel_size=3, stride=stride, padding=padding)
)
# 1x1 conv -> 3x3 conv branch -> 3x3 conv branch
self.branch3 = nn.Sequential(
BasicConv2d(in_channels, out5in_channels, kernel_size=1),
BasicConv2d(out5in_channels, out5_channels, kernel_size=3, padding=1),
BasicConv2d(out5_channels, out5_channels, kernel_size=3, stride=stride, padding=padding)
)
# 3x3 pool -> 1x1 conv branch
self.branch4 = nn.Sequential(
nn.MaxPool2d(3, stride=stride, padding=padding),
BasicConv2d(in_channels, outpool_channels, kernel_size=1)
)
def __init__(self, in_channels, b2_n1, b2_n3, b3_n1, b3_n1x7, b3_n7x1,
b3_n3):
super(Reduction_B_v4, self).__init__()
self.branch2 = nn.Sequential(
BasicConv2d(in_channels,
b2_n1,
kernel_size=1,
stride=1,
padding=0,
bias=False),
BasicConv2d(b2_n1,
b2_n3,
kernel_size=3,
stride=2,
padding=0,
bias=False))
self.branch3 = nn.Sequential(
BasicConv2d(in_channels,
b3_n1,
kernel_size=1,
stride=1,
padding=0,
bias=False),
BasicConv2d(b3_n1,
b3_n1x7,
kernel_size=(1, 7),
stride=(1, 1),
padding=(0, 3),
bias=False),
BasicConv2d(b3_n1x7,
b3_n7x1,
kernel_size=(7, 1),
stride=(1, 1),
padding=(3, 0),
bias=False),
BasicConv2d(b3_n7x1,
b3_n3,
kernel_size=3,
stride=2,
padding=0,
bias=False))
def __init__(self):
super(Stem_Res1, self).__init__()
self.stem = nn.Sequential(
BasicConv2d(3, 32, kernel_size=3, stride=2, padding=0, bias=False),
BasicConv2d(32, 32, kernel_size=3, stride=1, padding=0,
bias=False),
BasicConv2d(32, 64, kernel_size=3, stride=1, padding=1,
bias=False), nn.MaxPool2d(3, 2, 0),
BasicConv2d(64, 80, kernel_size=1, stride=1, padding=0,
bias=False),
BasicConv2d(80,
192,
kernel_size=3,
stride=1,
padding=0,
bias=False),
BasicConv2d(192,
256,
kernel_size=3,
stride=2,
padding=0,
bias=False))
def __init__(self, in_channels, b1, b2, b3_n1, b3_n1x3_3x1, b4_n1, b4_n1x3,
b4_n3x1, b4_n1x3_3x1):
super(InceptionC, self).__init__()
self.branch1 = nn.Sequential(
nn.AvgPool2d(3, 1, 1),
BasicConv2d(in_channels,
b1,
kernel_size=1,
stride=1,
padding=0,
bias=False))
self.branch2 = BasicConv2d(in_channels,
b2,
kernel_size=1,
stride=1,
padding=0,
bias=False)
self.branch3_1 = BasicConv2d(in_channels,
b3_n1,
kernel_size=1,
stride=1,
padding=0,
bias=False)
self.branch3_1x3 = BasicConv2d(b3_n1,
b3_n1x3_3x1,
kernel_size=(1, 3),
stride=(1, 1),
padding=(0, 1),
bias=False)
self.branch3_3x1 = BasicConv2d(b3_n1,
b3_n1x3_3x1,
kernel_size=(3, 1),
stride=(1, 1),
padding=(1, 0),
bias=False)
self.branch4_1 = nn.Sequential(
BasicConv2d(in_channels,
b4_n1,
kernel_size=1,
stride=1,
padding=0,
bias=False),
BasicConv2d(b4_n1,
b4_n1x3,
kernel_size=(1, 3),
stride=(1, 1),
padding=(0, 1),
bias=False),
BasicConv2d(b4_n1x3,
b4_n3x1,
kernel_size=(3, 1),
stride=(1, 1),
padding=(1, 0),
bias=False))
self.branch4_1x3 = BasicConv2d(b4_n3x1,
b4_n1x3_3x1,
kernel_size=(1, 3),
stride=(1, 1),
padding=(0, 1),
bias=False)
self.branch4_3x1 = BasicConv2d(b4_n3x1,
b4_n1x3_3x1,
kernel_size=(3, 1),
stride=(1, 1),
padding=(1, 0),
bias=False)
def __init__(self, in_channels, b1, b2, b3_n1, b3_n1x7, b3_n7x1, b4_n1,
b4_n1x7_1, b4_n7x1_1, b4_n1x7_2, b4_n7x1_2):
super(Inception_B, self).__init__()
self.branch1 = nn.Sequential(
nn.AvgPool2d(3, 1, 1),
BasicConv2d(in_channels,
b1,
kernel_size=1,
stride=1,
padding=0,
bias=False))
self.branch2 = BasicConv2d(in_channels,
b2,
kernel_size=1,
stride=1,
padding=0,
bias=False)
self.branch3 = nn.Sequential(
BasicConv2d(in_channels,
b3_n1,
kernel_size=1,
stride=1,
padding=0,
bias=False),
BasicConv2d(b3_n1,
b3_n1x7,
kernel_size=(1, 7),
stride=(1, 1),
padding=(0, 3),
bias=False),
BasicConv2d(b3_n1x7,
b3_n7x1,
kernel_size=(1, 7),
stride=(1, 1),
padding=(0, 3),
bias=False))
self.branch4 = nn.Sequential(
BasicConv2d(in_channels,
b4_n1,
kernel_size=1,
stride=1,
padding=0,
bias=False),
BasicConv2d(b4_n1,
b4_n1x7_1,
kernel_size=(1, 7),
stride=(1, 1),
padding=(0, 3),
bias=False),
BasicConv2d(b4_n1x7_1,
b4_n7x1_1,
kernel_size=(7, 1),
stride=(1, 1),
padding=(3, 0),
bias=False),
BasicConv2d(b4_n7x1_1,
b4_n1x7_2,
kernel_size=(1, 7),
stride=(1, 1),
padding=(0, 3),
bias=False),
BasicConv2d(b4_n1x7_2,
b4_n7x1_2,
kernel_size=(7, 1),
stride=(1, 1),
padding=(3, 0),
bias=False))
def __init__(self):
super(Stem_v4_Res2, self).__init__()
self.step1 = nn.Sequential(
BasicConv2d(3, 32, kernel_size=3, stride=2, padding=0, bias=False),
BasicConv2d(32, 32, kernel_size=3, stride=1, padding=0,
bias=False),
BasicConv2d(32, 64, kernel_size=3, stride=1, padding=1,
bias=False))
self.step2_pool = nn.MaxPool2d(3, 2, 0)
self.step2_conv = BasicConv2d(64,
96,
kernel_size=3,
stride=2,
padding=0,
bias=False)
self.step3_1 = nn.Sequential(
BasicConv2d(160,
64,
kernel_size=1,
stride=1,
padding=0,
bias=False),
BasicConv2d(64, 96, kernel_size=3, stride=1, padding=0,
bias=False))
self.step3_2 = nn.Sequential(
BasicConv2d(160,
64,
kernel_size=1,
stride=1,
padding=0,
bias=False),
BasicConv2d(64,
64,
kernel_size=(7, 1),
stride=(1, 1),
padding=(3, 0),
bias=False),
BasicConv2d(64,
64,
kernel_size=(1, 7),
stride=(1, 1),
padding=(0, 3),
bias=False),
BasicConv2d(64, 96, kernel_size=3, stride=1, padding=0,
bias=False))
self.step4_pool = nn.MaxPool2d(3, 2, 0)
self.step4_conv = BasicConv2d(192,
192,
kernel_size=3,
stride=2,
padding=0,
bias=False)