def __init__(self, num_classes=10):
super(VGG_SMALL_1W1A_AHTANHLayer, 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.nonlinear0 = alphaHtanhLayer()
self.conv1 = ir_1w32a.IRConv2d(128,
128,
kernel_size=3,
padding=1,
bias=False)
self.pooling = nn.MaxPool2d(kernel_size=2, stride=2)
self.bn1 = nn.BatchNorm2d(128)
self.nonlinear1 = alphaHtanhLayer()
# self.nonlinear = nn.ReLU(inplace=True)
# self.nonlinear = nn.Hardtanh(inplace=True)
self.conv2 = ir_1w32a.IRConv2d(128,
256,
kernel_size=3,
padding=1,
bias=False)
self.bn2 = nn.BatchNorm2d(256)
self.nonlinear2 = alphaHtanhLayer()
self.conv3 = ir_1w32a.IRConv2d(256,
256,
kernel_size=3,
padding=1,
bias=False)
self.bn3 = nn.BatchNorm2d(256)
self.nonlinear3 = alphaHtanhLayer()
self.conv4 = ir_1w32a.IRConv2d(256,
512,
kernel_size=3,
padding=1,
bias=False)
self.bn4 = nn.BatchNorm2d(512)
self.nonlinear4 = alphaHtanhLayer()
self.conv5 = ir_1w32a.IRConv2d(512,
512,
kernel_size=3,
padding=1,
bias=False)
self.bn5 = nn.BatchNorm2d(512)
self.nonlinear5 = alphaHtanhLayer()
self.fc = nn.Linear(512 * 4 * 4, self.num_classes)
self._initialize_weights()
def conv3x3Binary(in_planes, out_planes, stride=1):
"3x3 convolution with padding"
return ir_1w32a.IRConv2d(in_planes,
out_planes,
kernel_size=3,
stride=stride,
padding=1,
bias=False)
def __init__(self, in_planes, planes, stride=1, option='A'):
super(BasicBlock_1w32a, self).__init__()
self.conv1 = ir_1w32a.IRConv2d(in_planes, planes, kernel_size=3, stride=stride, padding=1, bias=False)
self.bn1 = nn.BatchNorm2d(planes)
self.conv2 = ir_1w32a.IRConv2d(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(
ir_1w32a.IRConv2d(in_planes, self.expansion * planes, kernel_size=1, stride=stride, bias=False),
nn.BatchNorm2d(self.expansion * planes)
)