Add files via upload

sew_resnet.py

@@ -0,0 +1,278 @@
+import torch
+import torch.nn as nn
+from spikingjelly.clock_driven import layer
+from spikingjelly.cext import neuron as cext_neuron
+__all__ = ['SEWResNet', 'sew_resnet18', 'sew_resnet34', 'sew_resnet50', 'sew_resnet101',
+           'sew_resnet152']
+
+def conv3x3(in_planes, out_planes, stride=1, groups=1, dilation=1):
+    """3x3 convolution with padding"""
+    return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride,
+                     padding=dilation, groups=groups, bias=False, dilation=dilation)
+
+
+def conv1x1(in_planes, out_planes, stride=1):
+    """1x1 convolution"""
+    return nn.Conv2d(in_planes, out_planes, kernel_size=1, stride=stride, bias=False)
+
+class LearnableSpike(nn.Module):
+    def __init__(self, out_chn):
+        super(LearnableSpike, self).__init__()
+        self.bias = nn.Parameter(torch.ones(1,1,out_chn,1,1), requires_grad=True)
+        #self.bias = nn.Parameter(torch.zeros(1,1,1,1,1), requires_grad=True)
+
+    def forward(self, x):
+        out = x * self.bias.expand_as(x)
+        return out
+
+class BasicBlock(nn.Module):
+    expansion = 1
+
+    def __init__(self, inplanes, planes, stride=1, downsample=None, groups=1,
+                 base_width=64, dilation=1, norm_layer=None, connect_f=None):
+        super(BasicBlock, self).__init__()
+        self.connect_f = connect_f
+        if norm_layer is None:
+            norm_layer = nn.BatchNorm2d
+        if groups != 1 or base_width != 64:
+            raise ValueError('SpikingBasicBlock only supports groups=1 and base_width=64')
+        if dilation > 1:
+            raise NotImplementedError("Dilation > 1 not supported in SpikingBasicBlock")
+
+        self.conv1 = layer.SeqToANNContainer(
+            conv3x3(inplanes, planes, stride),
+            norm_layer(planes)
+        )
+        self.sn1 = cext_neuron.MultiStepIFNode(detach_reset=True)
+
+        self.spike1 = LearnableSpike(planes)
+
+        self.conv2 = layer.SeqToANNContainer(
+            conv3x3(planes, planes),
+            norm_layer(planes)
+        )
+        self.downsample = downsample
+        self.stride = stride
+        self.sn2 = cext_neuron.MultiStepIFNode(detach_reset=True)
+
+        self.spike2 = LearnableSpike(planes)
+
+    def forward(self, x):
+        identity = x
+
+        out = self.sn1(self.conv1(x))
+        out = self.spike1(out)
+
+        out = self.sn2(self.conv2(out))
+        out = self.spike2(out)
+
+        if self.downsample is not None:
+            identity = self.downsample(x)
+
+        if self.connect_f == 'ADD':
+            out += identity
+        elif self.connect_f == 'AND':
+            out *= identity
+        elif self.connect_f == 'IAND':
+            out = identity * (1. - out)
+        else:
+            raise NotImplementedError(self.connect_f)
+
+        return out
+
+
+class Bottleneck(nn.Module):
+    expansion = 4
+    def __init__(self, inplanes, planes, stride=1, downsample=None, groups=1,
+                 base_width=64, dilation=1, norm_layer=None, connect_f=None):
+        super(Bottleneck, self).__init__()
+        self.connect_f = connect_f
+        if norm_layer is None:
+            norm_layer = nn.BatchNorm2d
+        width = int(planes * (base_width / 64.)) * groups
+        self.conv1 = layer.SeqToANNContainer(
+            conv1x1(inplanes, width),
+            norm_layer(width)
+        )
+        self.sn1 = cext_neuron.MultiStepIFNode(detach_reset=True)
+        self.spike1 = LearnableSpike(width)
+
+        self.conv2 = layer.SeqToANNContainer(
+            conv3x3(width, width, stride, groups, dilation),
+            norm_layer(width)
+        )
+        self.sn2 = cext_neuron.MultiStepIFNode(detach_reset=True)
+        self.spike2 = LearnableSpike(width)
+
+        self.conv3 = layer.SeqToANNContainer(
+            conv1x1(width, planes * self.expansion),
+            norm_layer(planes * self.expansion)
+        )
+        self.downsample = downsample
+        self.stride = stride
+        self.sn3 = cext_neuron.MultiStepIFNode(detach_reset=True)
+        self.spike3 = LearnableSpike(planes * self.expansion)
+
+    def forward(self, x):
+        identity = x
+
+        out = self.sn1(self.conv1(x))
+        out = self.spike1(out)
+
+        out = self.sn2(self.conv2(out))
+        out = self.spike2(out)
+
+        out = self.sn3(self.conv3(out))
+        out = self.spike3(out)
+
+        if self.downsample is not None:
+            identity = self.downsample(x)
+
+
+        if self.connect_f == 'ADD':
+            out += identity
+        elif self.connect_f == 'AND':
+            out *= identity
+        elif self.connect_f == 'IAND':
+            out = identity * (1. - out)
+        else:
+            raise NotImplementedError(self.connect_f)
+
+        return out
+def zero_init_blocks(net: nn.Module, connect_f: str):
+    for m in net.modules():
+        if isinstance(m, Bottleneck):
+            nn.init.constant_(m.conv3.module[1].weight, 0)
+            if connect_f == 'AND':
+                nn.init.constant_(m.conv3.module[1].bias, 1)
+        elif isinstance(m, BasicBlock):
+            nn.init.constant_(m.conv2.module[1].weight, 0)
+            if connect_f == 'AND':
+                nn.init.constant_(m.conv2.module[1].bias, 1)
+
+
+class SEWResNet(nn.Module):
+
+    def __init__(self, block, layers, num_classes=1000, zero_init_residual=False,
+                 groups=1, width_per_group=64, replace_stride_with_dilation=None,
+                 norm_layer=None, T=4, connect_f=None):
+        super(SEWResNet, self).__init__()
+        self.T = T
+        self.connect_f = connect_f
+        if norm_layer is None:
+            norm_layer = nn.BatchNorm2d
+        self._norm_layer = norm_layer
+
+        self.inplanes = 64
+        self.dilation = 1
+        if replace_stride_with_dilation is None:
+            replace_stride_with_dilation = [False, False, False]
+        if len(replace_stride_with_dilation) != 3:
+            raise ValueError("replace_stride_with_dilation should be None "
+                             "or a 3-element tuple, got {}".format(replace_stride_with_dilation))
+        self.groups = groups
+        self.base_width = width_per_group
+        self.conv1 = nn.Conv2d(3, self.inplanes, kernel_size=7, stride=2, padding=3,
+                               bias=False)
+        self.bn1 = norm_layer(self.inplanes)
+
+
+        self.sn1 = cext_neuron.MultiStepIFNode(detach_reset=True)
+        self.spike1 = LearnableSpike(self.inplanes)
+
+        self.maxpool = layer.SeqToANNContainer(nn.MaxPool2d(kernel_size=3, stride=2, padding=1))
+
+        self.layer1 = self._make_layer(block, 64, layers[0], connect_f=connect_f)
+        self.layer2 = self._make_layer(block, 128, layers[1], stride=2,
+                                       dilate=replace_stride_with_dilation[0], connect_f=connect_f)
+        self.layer3 = self._make_layer(block, 256, layers[2], stride=2,
+                                       dilate=replace_stride_with_dilation[1], connect_f=connect_f)
+        self.layer4 = self._make_layer(block, 512, layers[3], stride=2,
+                                       dilate=replace_stride_with_dilation[2], connect_f=connect_f)
+        self.avgpool = layer.SeqToANNContainer(nn.AdaptiveAvgPool2d((1, 1)))
+        self.fc = nn.Linear(512 * block.expansion, num_classes)
+
+        for m in self.modules():
+            if isinstance(m, nn.Conv2d):
+                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
+            elif isinstance(m, (nn.BatchNorm2d, nn.GroupNorm)):
+                nn.init.constant_(m.weight, 1)
+                nn.init.constant_(m.bias, 0)
+
+        if zero_init_residual:
+            zero_init_blocks(self, connect_f)
+
+    def _make_layer(self, block, planes, blocks, stride=1, dilate=False, connect_f=None):
+        norm_layer = self._norm_layer
+        downsample = None
+        previous_dilation = self.dilation
+        if dilate:
+            self.dilation *= stride
+            stride = 1
+        if stride != 1 or self.inplanes != planes * block.expansion:
+            downsample = nn.Sequential(
+                layer.SeqToANNContainer(
+                    conv1x1(self.inplanes, planes * block.expansion, stride),
+                    norm_layer(planes * block.expansion),
+                ),
+                cext_neuron.MultiStepIFNode(detach_reset=True)
+            )
+
+        layers = []
+        layers.append(block(self.inplanes, planes, stride, downsample, self.groups,
+                            self.base_width, previous_dilation, norm_layer, connect_f))
+        self.inplanes = planes * block.expansion
+        for _ in range(1, blocks):
+            layers.append(block(self.inplanes, planes, groups=self.groups,
+                                base_width=self.base_width, dilation=self.dilation,
+                                norm_layer=norm_layer, connect_f=connect_f))
+
+        return nn.Sequential(*layers)
+
+    def _forward_impl(self, x):
+        x = self.conv1(x)
+        x = self.bn1(x)
+        x.unsqueeze_(0)
+        x = x.repeat(self.T, 1, 1, 1, 1)
+        x = self.sn1(x)
+        x = self.spike1(x)
+        x = self.maxpool(x)
+        x = self.layer1(x)
+        x = self.layer2(x)
+        x = self.layer3(x)
+        x = self.layer4(x)
+
+        x = self.avgpool(x)
+        x = torch.flatten(x, 2)
+        return self.fc(x.mean(dim=0))
+
+    def forward(self, x):
+        return self._forward_impl(x)
+
+
+def _sew_resnet(block, layers, **kwargs):
+    model = SEWResNet(block, layers, **kwargs)
+    return model
+
+
+def sew_resnet18(**kwargs):
+    return _sew_resnet(BasicBlock, [2, 2, 2, 2], **kwargs)
+
+
+def sew_resnet34(**kwargs):
+    return _sew_resnet(BasicBlock, [3, 4, 6, 3], **kwargs)
+
+
+def sew_resnet50(**kwargs):
+    return _sew_resnet(Bottleneck, [3, 4, 6, 3], **kwargs)
+
+
+def sew_resnet101(**kwargs):
+    return _sew_resnet(Bottleneck, [3, 4, 23, 3], **kwargs)
+
+
+def sew_resnet152(**kwargs):
+    return _sew_resnet(Bottleneck, [3, 8, 36, 3], **kwargs)
+
+
+