def __init__(self):
super(Conv4_BinAct, self).__init__()
builder = get_builder()
self.convs = nn.Sequential(
builder.conv3x3(3, 64, first_layer=True),
builder.batchnorm(64),
BiRealAct(),
builder.conv3x3(64, 64),
nn.MaxPool2d((2, 2)),
builder.batchnorm(64),
BiRealAct(),
builder.conv3x3(64, 128),
builder.batchnorm(128),
BiRealAct(),
builder.conv3x3(128, 128),
nn.MaxPool2d((2, 2)),
builder.batchnorm(128),
BiRealAct(),
)
self.linear = nn.Sequential(
builder.conv1x1(32 * 32 * 8, 256),
builder.batchnorm(256),
BiRealAct(),
builder.conv1x1(256, 256),
builder.batchnorm(256),
BiRealAct(),
builder.conv1x1(256, 10),
)
def __init__(self):
super(Conv8Wide, self).__init__()
builder = get_builder()
self.convs = nn.Sequential(
builder.conv3x3(3, scale(64), first_layer=True),
nn.ReLU(),
builder.conv3x3(scale(64), scale(64)),
nn.ReLU(),
nn.MaxPool2d((2, 2)),
builder.conv3x3(scale(64), scale(128)),
nn.ReLU(),
builder.conv3x3(scale(128), scale(128)),
nn.ReLU(),
nn.MaxPool2d((2, 2)),
builder.conv3x3(scale(128), scale(256)),
nn.ReLU(),
builder.conv3x3(scale(256), scale(256)),
nn.ReLU(),
nn.MaxPool2d((2, 2)),
builder.conv3x3(scale(256), scale(512)),
nn.ReLU(),
builder.conv3x3(scale(512), scale(512)),
nn.ReLU(),
nn.MaxPool2d((2, 2))
)
self.linear = nn.Sequential(
builder.conv1x1(scale(512) * 2 * 2, scale(256)),
nn.ReLU(),
builder.conv1x1(scale(256), scale(256)),
nn.ReLU(),
builder.conv1x1(scale(256), 10),
)
def __init__(self):
super(Conv4Wide_BinAct, self).__init__()
builder = get_builder()
self.convs = nn.Sequential(
builder.conv3x3(3, scale(64), first_layer=True),
builder.batchnorm(scale(64)),
BiRealAct(),
builder.conv3x3(scale(64), scale(64)),
nn.MaxPool2d((2, 2)),
builder.batchnorm(scale(64)),
BiRealAct(),
builder.conv3x3(scale(64), scale(128)),
builder.batchnorm(scale(128)),
BiRealAct(),
builder.conv3x3(scale(128), scale(128)),
nn.MaxPool2d((2, 2)),
builder.batchnorm(scale(128)),
BiRealAct(),
)
self.linear = nn.Sequential(
builder.conv1x1(scale(128)*8*8, scale(256)),
builder.batchnorm(scale(256)),
BiRealAct(),
builder.conv1x1(scale(256), scale(256)),
builder.batchnorm(scale(256)),
BiRealAct(),
builder.conv1x1(scale(256), 10),
)
def __init__(self):
super(VGG_Small_noReLU_BinAct, self).__init__()
builder = get_builder()
self.convs = nn.Sequential(
builder.conv3x3(3, 128, first_layer=True),
#builder.batchnorm(128),
BiRealAct(),
builder.conv3x3(128, 128),
nn.MaxPool2d((2, 2)),
builder.batchnorm(128),
#nn.ReLU(),
BiRealAct(),
builder.conv3x3(128, 256),
#builder.batchnorm(256),
BiRealAct(),
builder.conv3x3(256, 256),
nn.MaxPool2d((2, 2)),
builder.batchnorm(256),
#nn.ReLU(),
BiRealAct(),
builder.conv3x3(256, 512),
#builder.batchnorm(512),
BiRealAct(),
builder.conv3x3(512, 512),
nn.MaxPool2d((2, 2)),
builder.batchnorm(512),
nn.ReLU(),
)
self.linear = nn.Sequential(
builder.conv1x1(512 * 4 * 4, 10),
)
def __init__(self):
super(Wide_VGG_Small, self).__init__()
builder = get_builder()
self.convs = nn.Sequential(
builder.conv3x3(3, scale(16), first_layer=True),
builder.batchnorm(scale(16)),
BiRealAct(),
builder.conv3x3(scale(16), scale(16)),
nn.MaxPool2d((2, 2)),
builder.batchnorm(scale(16)),
BiRealAct(),
builder.conv3x3(scale(16), scale(32)),
builder.batchnorm(scale(32)),
BiRealAct(),
builder.conv3x3(scale(32), scale(32)),
nn.MaxPool2d((2, 2)),
builder.batchnorm(scale(32)),
BiRealAct(),
builder.conv3x3(scale(32), scale(64)),
builder.batchnorm(scale(64)),
BiRealAct(),
builder.conv3x3(scale(64), scale(64)),
nn.MaxPool2d((2, 2)),
builder.batchnorm(scale(64)),
#BiRealAct(),
nn.ReLU(),
)
self.linear = nn.Sequential(
builder.conv1x1(scale(64) * 4 * 4, 10),
)
def __init__(self):
super(Conv8, self).__init__()
builder = get_builder()
self.convs = nn.Sequential(
builder.conv3x3(3, 64, first_layer=True),
nn.ReLU(),
builder.conv3x3(64, 64),
nn.ReLU(),
nn.MaxPool2d((2, 2)),
builder.conv3x3(64, 128),
nn.ReLU(),
builder.conv3x3(128, 128),
nn.ReLU(),
nn.MaxPool2d((2, 2)),
builder.conv3x3(128, 256),
nn.ReLU(),
builder.conv3x3(256, 256),
nn.ReLU(),
nn.MaxPool2d((2, 2)),
builder.conv3x3(256, 512),
nn.ReLU(),
builder.conv3x3(512, 512),
nn.ReLU(),
nn.MaxPool2d((2, 2))
)
self.linear = nn.Sequential(
builder.conv1x1(512 * 2 * 2, 256),
nn.ReLU(),
builder.conv1x1(256, 256),
nn.ReLU(),
builder.conv1x1(256, 10),
)
def vgg13_fc():
r"""VGG 13-layer model (configuration "B")
`"Very Deep Convolutional Networks For Large-Scale Image Recognition" <https://arxiv.org/pdf/1409.1556.pdf>`_
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
progress (bool): If True, displays a progress bar of the download to stderr
"""
return _vgg('B', False, get_builder())
def vgg11_bn_new_fc():
r"""VGG 11-layer model (configuration "A") with batch normalization
`"Very Deep Convolutional Networks For Large-Scale Image Recognition" <https://arxiv.org/pdf/1409.1556.pdf>`_
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
progress (bool): If True, displays a progress bar of the download to stderr
"""
return _vgg('A', True, get_builder())
def __init__(self):
super(FC, self).__init__()
builder = get_builder()
self.linear = nn.Sequential(
builder.conv1x1(28 * 28, 300, first_layer=True),
nn.ReLU(),
builder.conv1x1(300, 100),
nn.ReLU(),
builder.conv1x1(100, 10),
)
def __init__(self, block, num_blocks, num_classes):
super(ResNet, self).__init__()
self.builder = get_builder()
_outputs = [32, 64, 128]
self.in_planes = _outputs[0]
self.conv1 = self.builder.conv3x3(3, 32, stride=1, first_layer=True)
self.bn = nn.BatchNorm2d(_outputs[0])
self.layer1 = self._make_layer(block, 32, num_blocks[0], stride=1)
self.layer2 = self._make_layer(block, 64, num_blocks[1], stride=2)
self.layer3 = self._make_layer(block, 128, num_blocks[2], stride=2)
self.linear = self.builder.conv1x1(128, num_classes)
def resnet50():
block_cfg = [64, 128, 256, 512]
layer_cfg = [
64, 64, 64, 64, 64, 64, 128, 128, 128, 128, 128, 128, 128, 128, 256,
256, 256, 256, 256, 256, 256, 256, 256, 256, 256, 256, 512, 512, 512,
512, 512, 512
]
return ResNet(get_builder(),
Bottleneck, [3, 4, 6, 3],
block_cfg=block_cfg,
layer_cfg=layer_cfg,
num_classes=1000)
def __init__(self):
super(Conv2, self).__init__()
builder = get_builder()
self.convs = nn.Sequential(
builder.conv3x3(3, 64, first_layer=True),
nn.ReLU(),
builder.conv3x3(64, 64),
nn.ReLU(),
nn.MaxPool2d((2, 2)),
)
self.linear = nn.Sequential(
builder.conv1x1(64 * 16 * 16, 256),
nn.ReLU(),
builder.conv1x1(256, 256),
nn.ReLU(),
builder.conv1x1(256, 10),
)
def mobilenet_v3_small(**kwargs):
"""
Constructs a MobileNetV3-Small model
"""
cfgs = [
# k, t, c, SE, HS, s
[3, 1, 16, 1, 0, 2],
[3, 4.5, 24, 0, 0, 2],
[3, 3.67, 24, 0, 0, 1],
[5, 4, 40, 1, 1, 2],
[5, 6, 40, 1, 1, 1],
[5, 6, 40, 1, 1, 1],
[5, 3, 48, 1, 1, 1],
[5, 3, 48, 1, 1, 1],
[5, 6, 96, 1, 1, 2],
[5, 6, 96, 1, 1, 1],
[5, 6, 96, 1, 1, 1],
]
return MobileNetV3(get_builder(), cfgs, mode='small', **kwargs)
def mobilenet_v3_large(**kwargs):
"""
Constructs a MobileNetV3-Large model
"""
cfgs = [
# k, t, c, SE, HS, s
[3, 1, 16, 0, 0, 1],
[3, 4, 24, 0, 0, 2],
[3, 3, 24, 0, 0, 1],
[5, 3, 40, 1, 0, 2],
[5, 3, 40, 1, 0, 1],
[5, 3, 40, 1, 0, 1],
[3, 6, 80, 0, 1, 2],
[3, 2.5, 80, 0, 1, 1],
[3, 2.3, 80, 0, 1, 1],
[3, 2.3, 80, 0, 1, 1],
[3, 6, 112, 1, 1, 1],
[3, 6, 112, 1, 1, 1],
[5, 6, 160, 1, 1, 2],
[5, 6, 160, 1, 1, 1],
[5, 6, 160, 1, 1, 1]
]
return MobileNetV3(get_builder(), cfgs, mode='large', **kwargs)
def __init__(self):
super(MobileNetV1, self).__init__()
builder = get_builder()
def conv_bn(inp, oup, stride):
return nn.Sequential(
builder.conv2d(inp, oup, 3, stride, 1, bias=False),
builder.batchnorm(oup), nn.ReLU(inplace=True))
def conv_dw(inp, oup, stride):
return nn.Sequential(
builder.conv2d(inp, inp, 3, stride, 1, groups=inp, bias=False),
builder.batchnorm(inp),
nn.ReLU(inplace=True),
builder.conv2d(inp, oup, 1, 1, 0, bias=False),
builder.batchnorm(oup),
nn.ReLU(inplace=True),
)
self.model = nn.Sequential(
conv_bn(3, 32, 2),
conv_dw(32, 64, 1),
conv_dw(64, 128, 2),
conv_dw(128, 128, 1),
conv_dw(128, 256, 2),
conv_dw(256, 256, 1),
conv_dw(256, 512, 2),
conv_dw(512, 512, 1),
conv_dw(512, 512, 1),
conv_dw(512, 512, 1),
conv_dw(512, 512, 1),
conv_dw(512, 512, 1),
conv_dw(512, 1024, 2),
conv_dw(1024, 1024, 1),
nn.AvgPool2d(7),
)
self.fc = builder.conv1x1(1024, 1000)
def __init__(self):
super(Conv6_BNN, self).__init__()
builder = get_builder()
self.convs = nn.Sequential(
builder.conv3x3(3, 128, first_layer=True),
builder.batchnorm(128),
BinAct(),
builder.conv3x3(128, 128),
nn.MaxPool2d((2, 2)),
builder.batchnorm(128),
BinAct(),
builder.conv3x3(128, 256),
builder.batchnorm(256),
BinAct(),
builder.conv3x3(256, 256),
nn.MaxPool2d((2, 2)),
builder.batchnorm(256),
BinAct(),
builder.conv3x3(256, 512),
builder.batchnorm(512),
BinAct(),
builder.conv3x3(512, 512),
nn.MaxPool2d((2, 2)),
builder.batchnorm(512),
BinAct()
)
self.linear = nn.Sequential(
builder.conv1x1(512 * 4 * 4, 1024),
builder.batchnorm(1024),
BinAct(),
builder.conv1x1(1024, 1024),
builder.batchnorm(1024),
BinAct(),
builder.conv1x1(1024, 10)
)
def cResNet34_BinAct():
return ResNet_BinAct(get_builder(), BasicBlock_BinAct, [3, 4, 6, 3])
def cWideResNet18_3_BinAct():
return WideResNet_BinAct(get_builder(),
BasicBlock_BinAct, [2, 2, 2, 2],
widen_factor=3)
def cResNet18_BinAct():
return ResNet_BinAct(get_builder(), BasicBlock_BinAct, [2, 2, 2, 2])
def cResNet110():
return SmallResNet(get_builder(), BasicBlock, [18, 18, 18])
def cResNet56_BinAct():
return SmallResNet_BinAct(get_builder(), BasicBlock_BinAct, [9, 9, 9])
def cResNet32_BinAct():
return SmallResNet_BinAct(get_builder(), BasicBlock_BinAct, [5, 5, 5])
def cResNet20():
return SmallResNet(get_builder(), BasicBlock, [3, 3, 3])
def cWideResNeXt18_2_BinAct_small():
return WideResNeXt_BinAct(get_builder(),
Bottleneck2, [1, 2, 6, 2], [4, 4, 8, 8],
widen_factor=2)
def cResNet20_BinAct():
return SmallResNet_BinAct(get_builder(), BasicBlock_BinAct, [3, 3, 3])
def cResNet32():
return SmallResNet(get_builder(), BasicBlock, [5, 5, 5])
def cResNet44_BinAct():
return SmallResNet_BinAct(get_builder(), BasicBlock_BinAct, [7, 7, 7])
def cResNet44():
return SmallResNet(get_builder(), BasicBlock, [7, 7, 7])
def cResNet110_BinAct():
return SmallResNet_BinAct(get_builder(), BasicBlock_BinAct, [18, 18, 18])
def cResNet56():
return SmallResNet(get_builder(), BasicBlock, [9, 9, 9])