def sepconv2d(cin, cout=None, ksize=3, stride=1, padding=None, affine=True):
if cout is None: cout = cin
if padding is None: padding = ksize // 2
layer = nn.Sequential(
nn.ReLU(inplace=False),
init_default(
nn.Conv2d(cin,
cin,
ksize,
stride=stride,
padding=padding,
groups=cin,
bias=False), nn.init.kaiming_normal_),
init_default(nn.Conv2d(cin, cin, 1, padding=0, bias=False),
nn.init.kaiming_normal_),
nn.BatchNorm2d(cin, affine=affine), nn.ReLU(inplace=False),
init_default(
nn.Conv2d(cin,
cin,
ksize,
stride=1,
padding=padding,
groups=cin,
bias=False), nn.init.kaiming_normal_),
init_default(nn.Conv2d(cin, cout, 1, padding=0, bias=False),
nn.init.kaiming_normal_),
nn.BatchNorm2d(cout, affine=affine))
return layer
def __init__(self, num_classes=10):
super(ConvNet, self).__init__()
self.layer1 = nn.Sequential(
nn.Conv2d(1, 16, kernel_size=5, stride=1, padding=2),
nn.BatchNorm2d(16),
nn.ReLU(),
nn.MaxPool2d(kernel_size=2, stride=2))
self.layer2 = nn.Sequential(
nn.Conv2d(16, 32, kernel_size=5, stride=1, padding=2),
nn.BatchNorm2d(32),
nn.ReLU(),
nn.MaxPool2d(kernel_size=2, stride=2))
self.fc = nn.Linear(7 * 7 * 32, num_classes)
def conv2dpool(cin, cout, pool_type, bn=NormType.Batch):
assert pool_type in ['avg', 'max']
if pool_type == 'max':
return nn.Sequential(
nn.MaxPool2d(2, stride=2),
init_default(nn.Conv2d(cin, cout, 1, bias=False),
nn.init.kaiming_normal_),
batchnorm_2d(cout, norm_type=bn))
if pool_type == 'avg':
return nn.Sequential(
nn.AvgPool2d(2, stride=2, ceil_mode=True, count_include_pad=False),
init_default(nn.Conv2d(cin, cout, 1, bias=False),
nn.init.kaiming_normal_),
batchnorm_2d(cout, norm_type=bn))
def conv2d(cin,
cout=None,
ksize=3,
stride=1,
padding=None,
dilation=None,
groups=None,
use_relu=True,
use_bn=True,
bn=NormType.Batch,
bias=False):
if cout is None: cout = cin
if padding is None: padding = ksize // 2
if dilation is None: dilation = 1
if groups is None: groups = 1
layer = [
init_default(
nn.Conv2d(cin,
cout,
ksize,
stride=stride,
padding=padding,
dilation=dilation,
groups=groups,
bias=bias), nn.init.kaiming_normal_)
]
if use_bn: layer.append(batchnorm_2d(cout, norm_type=bn))
if use_relu: layer.append(relu(True))
return nn.Sequential(*layer)
def conv1x1(in_planes, out_planes, stride=1):
"""1x1 convolution"""
return nn.Conv2d(in_planes,
out_planes,
kernel_size=1,
stride=stride,
bias=False)
def conv3x3(in_planes, out_planes, stride=1):
"""3x3 convolution with padding"""
return nn.Conv2d(in_planes,
out_planes,
kernel_size=3,
stride=stride,
padding=1,
bias=False)
def two_conv_pool(self, in_channels, f1, f2):
s = nn.Sequential(
nn.Conv2d(in_channels, f1, kernel_size=3, stride=1, padding=1),
nn.BatchNorm2d(f1),
nn.ReLU(inplace=True),
nn.Conv2d(f1, f2, kernel_size=3, stride=1, padding=1),
nn.BatchNorm2d(f2),
nn.ReLU(inplace=True),
nn.MaxPool2d(kernel_size=2, stride=2),
)
for m in s.children():
if isinstance(m, nn.Conv2d):
n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
m.weight.data.normal_(0, math.sqrt(2. / n))
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
return s
def stem_blk(cin,
cout=None,
ksize=3,
stride=1,
use_relu=True,
use_bn=True,
bn=NormType.Batch,
bias=False,
pool='avg'):
if cout is None: cout = cin
padding = ksize // 2
layer = [
init_default(
nn.Conv2d(cin,
cout,
ksize,
stride=stride,
padding=padding,
bias=bias), nn.init.kaiming_normal_)
]
if use_bn: layer.append(batchnorm_2d(cout, norm_type=bn))
if use_relu: layer.append(relu(True))
if pool == 'max': layer.append(nn.MaxPool2d(2, stride=2))
if pool == 'avg':
layer.append(
nn.AvgPool2d(2, stride=2, ceil_mode=True, count_include_pad=False))
layer.append(
init_default(
nn.Conv2d(cout,
cout * 2,
ksize,
stride=stride,
padding=padding,
bias=bias), nn.init.kaiming_normal_))
if use_bn: layer.append(batchnorm_2d(cout * 2, norm_type=bn))
if use_relu: layer.append(relu(True))
if pool == 'max': layer.append(nn.MaxPool2d(2, stride=2))
if pool == 'avg':
layer.append(
nn.AvgPool2d(2, stride=2, ceil_mode=True, count_include_pad=False))
return nn.Sequential(*layer)
def __init__(self,
block,
layers,
num_classes=10,
zero_init_residual=False):
super(MyResNet, self).__init__()
self.inplanes = 64
self.conv1 = nn.Conv2d(1,
64,
kernel_size=7,
stride=2,
padding=3,
bias=False)
self.bn1 = nn.BatchNorm2d(64)
self.relu = nn.ReLU(inplace=True)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
self.layer1 = self._make_layer(block, 64, layers[0])
self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
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.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
# Zero-initialize the last BN in each residual branch,
# so that the residual branch starts with zeros, and each residual block behaves like an identity.
# This improves the model by 0.2~0.3% according to https://arxiv.org/abs/1706.02677
if zero_init_residual:
for m in self.modules():
if isinstance(m, Bottleneck):
nn.init.constant_(m.bn3.weight, 0)
elif isinstance(m, BasicBlock):
nn.init.constant_(m.bn2.weight, 0)
self.classifier = nn.Sequential(
nn.Dropout(p=0.5),
nn.Linear(512 * block.expansion, 256),
nn.BatchNorm1d(256),
nn.ReLU(inplace=True),
nn.Dropout(p=0.5),
nn.Linear(256, num_classes),
)
def conv(self, ni, nf):
return nn.Conv2d(ni, nf, kernel_size=3, stride=2, padding=1)
def stem(self):
return nn.Sequential(init_default(nn.Conv2d(3, self.channels, 3, padding=1, bias=False),
nn.init.kaiming_normal_),
nn.BatchNorm2d(self.channels))