def __init__(self, in_planes, planes, stride=1):
super(Bottleneck_1w1a, self).__init__()
self.conv1 = Layer.BNNConv2d_1w1a(in_planes,
planes,
kernel_size=1,
bias=False)
self.bn1 = nn.BatchNorm2d(planes)
self.conv2 = Layer.BNNConv2d_1w1a(planes,
planes,
kernel_size=3,
stride=stride,
padding=1,
bias=False)
self.bn2 = nn.BatchNorm2d(planes)
self.conv3 = Layer.BNNConv2d_1w1a(planes,
self.expansion * planes,
kernel_size=1,
bias=False)
self.bn3 = nn.BatchNorm2d(self.expansion * planes)
self.shortcut = nn.Sequential()
if stride != 1 or in_planes != self.expansion * planes:
self.shortcut = nn.Sequential(
Layer.BNNConv2d_1w1a(in_planes,
self.expansion * planes,
kernel_size=1,
stride=stride,
bias=False),
nn.BatchNorm2d(self.expansion * planes))
def __init__(self, in_planes, out_planes, stride=1):
super(Block_1w1a, self).__init__()
self.conv1 = Layer.BNNConv2d_1w1a(in_planes,
in_planes,
kernel_size=3,
stride=stride,
padding=1,
groups=in_planes,
bias=False)
self.bn1 = nn.BatchNorm2d(in_planes)
self.conv2 = Layer.BNNConv2d_1w1a(in_planes,
out_planes,
kernel_size=1,
stride=1,
padding=0,
bias=False)
self.bn2 = nn.BatchNorm2d(out_planes)
def __init__(self, num_classes=10):
super(VGG_SMALL_1W1A_MOVE, self).__init__()
self.num_classes = num_classes
self.conv0 = nn.Conv2d(3, 128, kernel_size=3, padding=1, bias=False)
self.bn0 = nn.BatchNorm2d(128)
self.move1 = MoveBlock(128)
self.conv1 = Layer.BNNConv2d_1w1a(128,
128,
kernel_size=3,
padding=1,
bias=False)
self.pooling = nn.MaxPool2d(kernel_size=2, stride=2)
self.bn1 = nn.BatchNorm2d(128)
# self.nonlinear = nn.ReLU(inplace=True)
# self.nonlinear = nn.Hardtanh(inplace=True)
self.move2 = MoveBlock(128)
self.conv2 = Layer.BNNConv2d_1w1a(128,
256,
kernel_size=3,
padding=1,
bias=False)
self.bn2 = nn.BatchNorm2d(256)
self.move3 = MoveBlock(256)
self.conv3 = Layer.BNNConv2d_1w1a(256,
256,
kernel_size=3,
padding=1,
bias=False)
self.bn3 = nn.BatchNorm2d(256)
self.move4 = MoveBlock(256)
self.conv4 = Layer.BNNConv2d_1w1a(256,
512,
kernel_size=3,
padding=1,
bias=False)
self.bn4 = nn.BatchNorm2d(512)
self.move5 = MoveBlock(512)
self.conv5 = Layer.BNNConv2d_1w1a(512,
512,
kernel_size=3,
padding=1,
bias=False)
self.bn5 = nn.BatchNorm2d(512)
self.fc = nn.Linear(512 * 4 * 4, self.num_classes)
self._initialize_weights()
def __init__(self, num_classes=10):
super(VGG_SMALL_1W1A_DENSE, self).__init__()
self.num_classes = num_classes
self.conv0 = nn.Conv2d(3, 128, kernel_size=3, padding=1, bias=False)
self.bn0 = nn.BatchNorm2d(128)
self.conv1 = Layer.BNNConv2d_1w1a(128,
128,
kernel_size=3,
padding=1,
bias=False)
self.pooling = nn.MaxPool2d(kernel_size=2, stride=2)
self.bn1 = nn.BatchNorm2d(128)
# self.nonlinear = nn.ReLU(inplace=True)
# self.nonlinear = nn.Hardtanh(inplace=True)
self.conv2 = Layer.BNNConv2d_1w1a(128,
256,
kernel_size=3,
padding=1,
bias=False)
self.bn2 = nn.BatchNorm2d(256)
self.conv3 = Layer.BNNConv2d_1w1a(256,
256,
kernel_size=3,
padding=1,
bias=False)
self.bn3 = nn.BatchNorm2d(256)
self.conv4 = Layer.BNNConv2d_1w1a(256,
512,
kernel_size=3,
padding=1,
bias=False)
self.bn4 = nn.BatchNorm2d(512)
self.conv5 = Layer.BNNConv2d_1w1a(512,
512,
kernel_size=3,
padding=1,
bias=False)
self.bn5 = nn.BatchNorm2d(512)
self.fc = Layer.BNNDense_1w1a(512 * 4 * 4, self.num_classes)
self.bn6 = nn.BatchNorm1d(self.num_classes)
self.scaleshift = ScaleAndShift(self.num_classes)
self._initialize_weights()
def __init__(self, in_planes, planes, stride=1, option='A'):
super(BasicBlock_1w1a, self).__init__()
self.conv1 = Layer.BNNConv2d_1w1a(in_planes, planes, kernel_size=3, stride=stride, padding=1, bias=False)
self.bn1 = nn.BatchNorm2d(planes)
self.conv2 = Layer.BNNConv2d_1w1a(planes, planes, kernel_size=3, stride=1, padding=1, bias=False)
self.bn2 = nn.BatchNorm2d(planes)
self.shortcut = nn.Sequential()
if stride != 1 or in_planes != planes:
if option == 'A':
"""
For CIFAR10 ResNet paper uses option A.
"""
self.shortcut = LambdaLayer(lambda x:
F.pad(x[:, :, ::2, ::2], (0, 0, 0, 0, planes//4, planes//4), "constant", 0))
elif option == 'B':
self.shortcut = nn.Sequential(
Layer.BNNConv2d_1w1a(in_planes, self.expansion * planes, kernel_size=1, stride=stride, bias=False),
nn.BatchNorm2d(self.expansion * planes)
)
def __init__(self, inp, oup, stride, expand_ratio):
super(InvertedResidual_1w1a, self).__init__()
self.stride = stride
assert stride in [1, 2]
hidden_dim = int(inp * expand_ratio)
self.use_res_connect = self.stride == 1 and inp == oup
if expand_ratio == 1:
self.conv = nn.Sequential(
# dw
Layer.BNNConv2d_1w1a(hidden_dim,
hidden_dim,
3,
stride,
1,
groups=hidden_dim,
bias=False),
nn.BatchNorm2d(hidden_dim),
# pw-linear
Layer.BNNConv2d_1w1a(hidden_dim, oup, 1, 1, 0, bias=False),
nn.BatchNorm2d(oup),
)
else:
self.conv = nn.Sequential(
# pw
Layer.BNNConv2d_1w1a(inp, hidden_dim, 1, 1, 0, bias=False),
nn.BatchNorm2d(hidden_dim),
# dw
Layer.BNNConv2d_1w1a(hidden_dim,
hidden_dim,
3,
stride,
1,
groups=hidden_dim,
bias=False),
nn.BatchNorm2d(hidden_dim),
# pw-linear
Layer.BNNConv2d_1w1a(hidden_dim, oup, 1, 1, 0, bias=False),
nn.BatchNorm2d(oup),
)
def conv_1x1_bn_1w1a(inp, oup):
return nn.Sequential(Layer.BNNConv2d_1w1a(inp, oup, 1, 1, 0, bias=False),
nn.BatchNorm2d(oup))