def __init__(self, tau, T, v_threshold=1.0, v_reset=0.0):
super().__init__()
self.T = T
self.static_conv = nn.Sequential(
nn.Conv2d(1, 128, kernel_size=3, padding=1, bias=False),
nn.BatchNorm2d(128),
)
self.conv = nn.Sequential(
neuron.IFNode(v_threshold=v_threshold, v_reset=v_reset, surrogate_function=surrogate.ATan()),
nn.MaxPool2d(2, 2), # 14 * 14
nn.Conv2d(128, 128, kernel_size=3, padding=1, bias=False),
nn.BatchNorm2d(128),
neuron.IFNode(v_threshold=v_threshold, v_reset=v_reset, surrogate_function=surrogate.ATan()),
nn.MaxPool2d(2, 2) # 7 * 7
)
self.fc = nn.Sequential(
nn.Flatten(),
layer.Dropout(0.7),
nn.Linear(128 * 7 * 7, 128 * 3 * 3, bias=False),
neuron.LIFNode(tau=tau, v_threshold=v_threshold, v_reset=v_reset, surrogate_function=surrogate.ATan()),
layer.Dropout(0.7),
nn.Linear(128 * 3 * 3, 128, bias=False),
neuron.LIFNode(tau=tau, v_threshold=v_threshold, v_reset=v_reset, surrogate_function=surrogate.ATan()),
nn.Linear(128, 10, bias=False),
neuron.LIFNode(tau=tau, v_threshold=v_threshold, v_reset=v_reset, surrogate_function=surrogate.ATan()),
)
def __init__(self, channels: int = 128):
super().__init__()
conv = []
conv.extend(SJSNN.conv3x3(2, channels))
conv.append(nn.MaxPool2d(2, 2))
for i in range(4):
conv.extend(SJSNN.conv3x3(channels, channels))
conv.append(nn.MaxPool2d(2, 2))
self.conv = nn.Sequential(*conv)
self.fc = nn.Sequential(
nn.Flatten(), layer.Dropout(0.5),
nn.Linear(channels * 4 * 4, channels * 2 * 2, bias=False),
neuron.LIFNode(tau=2.0,
surrogate_function=surrogate.ATan(),
detach_reset=True), layer.Dropout(0.5),
nn.Linear(channels * 2 * 2, 110, bias=False),
neuron.LIFNode(tau=2.0,
surrogate_function=surrogate.ATan(),
detach_reset=True))
self.vote = VotingLayer(10)
def __init__(self, in_c = 1,nf = [16,32], ks=[5,5], T = 10, v_threshold = 1.0, v_reset = 0.0, dropout=0.5, use_softmax = False, tau=2.0, lif=False): super().__init__() self.T = T self.use_softmax = use_softmax self.static_conv = nn.Sequential( nn.Conv2d(in_c, nf[0], kernel_size=ks[0], bias=False), nn.BatchNorm2d(nf[0]) ) self.dim1 = (28 - ks[0] + 1) // 2 self.conv = nn.Sequential( neuron.IFNode(v_threshold=v_threshold, v_reset=v_reset, surrogate_function=surrogate.ATan(), detach_reset=True), nn.MaxPool2d(2, 2), nn.Conv2d(nf[0], nf[1], kernel_size=ks[1], bias=False), nn.BatchNorm2d(nf[1]), neuron.IFNode(v_threshold=v_threshold, v_reset=v_reset, surrogate_function=surrogate.ATan(), detach_reset=True), nn.MaxPool2d(2, 2) ) if lif: self.conv[0] = neuron.LIFNode(tau=tau, v_threshold=v_threshold, v_reset=v_reset, surrogate_function=surrogate.ATan()) self.conv[4] = neuron.LIFNode(tau=tau, v_threshold=v_threshold, v_reset=v_reset, surrogate_function=surrogate.ATan()) self.dim2 = (self.dim1 - ks[1] + 1) // 2 self.fc = nn.Sequential( nn.Flatten(), layer.Dropout(dropout), nn.Linear(nf[1]*self.dim2**2, 10, bias=False), neuron.IFNode(v_threshold=v_threshold, v_reset=v_reset, surrogate_function=surrogate.ATan(), detach_reset=True) ) if lif: self.fc[-1] = neuron.LIFNode(tau=tau, v_threshold=v_threshold, v_reset=v_reset, surrogate_function=surrogate.ATan()) if self.use_softmax: self.softmax = nn.Softmax(dim=1)
def __init__(self,
T=8,
v_threshold=1.0,
v_reset=0.0,
tau=2.0,
surrogate_function=surrogate.ATan()):
super().__init__()
self.train_times = 0
self.epochs = 0
self.max_test_acccuracy = 0
self.T = T
self.static_conv = nn.Sequential(
nn.Conv2d(3, 256, kernel_size=3, padding=1, bias=False),
nn.BatchNorm2d(256),
)
self.conv = nn.Sequential(
neuron.LIFNode(v_threshold=v_threshold,
v_reset=v_reset,
tau=tau,
surrogate_function=surrogate_function,
detach_reset=True),
nn.Conv2d(256, 256, kernel_size=3, padding=1, bias=False),
nn.BatchNorm2d(256),
neuron.LIFNode(v_threshold=v_threshold,
v_reset=v_reset,
tau=tau,
surrogate_function=surrogate_function,
detach_reset=True),
nn.Conv2d(256, 256, kernel_size=3, padding=1, bias=False),
nn.BatchNorm2d(256),
neuron.LIFNode(v_threshold=v_threshold,
v_reset=v_reset,
tau=tau,
surrogate_function=surrogate_function,
detach_reset=True),
nn.MaxPool2d(2, 2), # 16 * 16
nn.Conv2d(256, 256, kernel_size=3, padding=1, bias=False),
nn.BatchNorm2d(256),
neuron.LIFNode(v_threshold=v_threshold,
v_reset=v_reset,
tau=tau,
surrogate_function=surrogate_function,
detach_reset=True),
nn.Conv2d(256, 256, kernel_size=3, padding=1, bias=False),
nn.BatchNorm2d(256),
neuron.LIFNode(v_threshold=v_threshold,
v_reset=v_reset,
tau=tau,
surrogate_function=surrogate_function,
detach_reset=True),
nn.Conv2d(256, 256, kernel_size=3, padding=1, bias=False),
nn.BatchNorm2d(256),
neuron.LIFNode(v_threshold=v_threshold,
v_reset=v_reset,
tau=tau,
surrogate_function=surrogate_function,
detach_reset=True),
nn.MaxPool2d(2, 2) # 8 * 8
)
self.fc = nn.Sequential(
nn.Flatten(), layer.Dropout(0.5),
nn.Linear(256 * 8 * 8, 128 * 4 * 4, bias=False),
neuron.LIFNode(v_threshold=v_threshold,
v_reset=v_reset,
tau=tau,
surrogate_function=surrogate_function,
detach_reset=True),
nn.Linear(128 * 4 * 4, 100, bias=False),
neuron.LIFNode(v_threshold=v_threshold,
v_reset=v_reset,
tau=tau,
surrogate_function=surrogate_function,
detach_reset=True))
self.boost = nn.AvgPool1d(10, 10)
def __init__(self): super().__init__() self.train_epoch = 0 self.cnn11 = nn.Conv2d(in_channels=3, out_channels=64, kernel_size=3, stride=1, padding=1, bias=False) self.lif11 = nn.Sequential( LIFNode(), layer.Dropout(0.25) ) self.avgpool1 = nn.AvgPool2d(kernel_size=2) self.if1 = IFNode() self.cnn21 = nn.Conv2d(in_channels=64, out_channels=128, kernel_size=3, stride=1, padding=1, bias=False) self.lif21 = nn.Sequential( LIFNode(), layer.Dropout(0.25) ) self.cnn22 = nn.Conv2d(in_channels=128, out_channels=128, kernel_size=3, stride=1, padding=1, bias=False) self.shortcut1 = nn.Sequential(nn.Conv2d(64, 128, kernel_size=1, stride=1, bias=False),) self.lif2 = nn.Sequential( LIFNode(), layer.Dropout(0.25) ) self.cnn31 = nn.Conv2d(in_channels=128, out_channels=256, kernel_size=3, stride=1, padding=1, bias=False) self.lif31 = nn.Sequential( LIFNode(), layer.Dropout(0.25) ) self.cnn32 = nn.Conv2d(in_channels=256, out_channels=256, kernel_size=3, stride=2, padding=1, bias=False) self.shortcut2 = nn.Sequential(nn.Conv2d(128, 256, kernel_size=1, stride=2, bias=False),) self.lif3 = nn.Sequential( LIFNode(), layer.Dropout(0.25) ) self.cnn41 = nn.Conv2d(in_channels=256, out_channels=512, kernel_size=3, stride=1, padding=1, bias=False) self.lif41 = nn.Sequential( LIFNode(), layer.Dropout(0.25) ) self.cnn42 = nn.Conv2d(in_channels=512, out_channels=512, kernel_size=3, stride=1, padding=1, bias=False) self.shortcut3 = nn.Sequential(nn.Conv2d(256, 512, kernel_size=1, stride=1, bias=False),) self.lif4 = nn.Sequential( LIFNode(), layer.Dropout(0.25) ) self.cnn51 = nn.Conv2d(in_channels=512, out_channels=512, kernel_size=3, stride=1, padding=1, bias=False) self.lif51 = nn.Sequential( LIFNode(), layer.Dropout(0.25) ) self.cnn52 = nn.Conv2d(in_channels=512, out_channels=512, kernel_size=3, stride=2, padding=1, bias=False) self.shortcut4 = nn.Sequential(nn.AvgPool2d(kernel_size=(1, 1), stride=(2, 2), padding=(0, 0))) self.lif5 = nn.Sequential( LIFNode(), layer.Dropout(0.25) ) self.fc0 = nn.Linear(512 * 4 * 4, 1024, bias=False) self.lif6 = nn.Sequential( LIFNode(), layer.Dropout(0.25) ) self.fc1 = nn.Linear(1024, 10, bias=False) self.lif_out = LIFNode(fire=False) for m in self.modules(): if isinstance(m, nn.Conv2d): n = m.kernel_size[0] * m.kernel_size[1] * m.in_channels variance1 = math.sqrt(1.0 / n) m.weight.data.normal_(0, variance1) elif isinstance(m, nn.Linear): size = m.weight.size() fan_in = size[1] variance2 = math.sqrt(1.0 / fan_in) m.weight.data.normal_(0.0, variance2)