def __init__(self,
inplanes,
planes,
stride=1,
downsample=None,
using_moving_average=False,
using_bn=True):
super(Bottleneck, self).__init__()
self.sn1 = sn.SwitchNorm2d(inplanes,
using_moving_average=using_moving_average,
using_bn=using_bn)
self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False)
self.sn2 = sn.SwitchNorm2d(planes,
using_moving_average=using_moving_average,
using_bn=using_bn)
self.conv2 = nn.Conv2d(planes,
planes,
kernel_size=3,
stride=stride,
padding=1,
bias=False)
self.sn3 = sn.SwitchNorm2d(planes,
using_moving_average=using_moving_average,
using_bn=using_bn)
self.conv3 = nn.Conv2d(planes, planes * 4, kernel_size=1, bias=False)
self.relu = nn.ReLU(inplace=True)
self.downsample = downsample
self.stride = stride
def __init__(self, inplanes, planes, stride=1, downsample=None, using_moving_average=True, using_bn=True, last_gamma=True):
super(BasicBlock, self).__init__()
self.conv1 = conv3x3(inplanes, planes, stride)
self.sn1 = sn.SwitchNorm2d(planes, using_moving_average=using_moving_average, using_bn=using_bn)
self.relu = nn.ReLU(inplace=True)
self.conv2 = conv3x3(planes, planes)
self.sn2 = sn.SwitchNorm2d(planes, using_moving_average=using_moving_average, using_bn=using_bn, last_gamma=last_gamma)
self.downsample = downsample
self.stride = stride
def __init__(self,
block,
layers,
num_classes=1000,
using_moving_average=True,
using_bn=True):
self.inplanes = 64
self.using_moving_average = using_moving_average
self.using_bn = using_bn
super(ResNetV2SN, self).__init__()
self.conv1 = nn.Conv2d(3,
64,
kernel_size=7,
stride=2,
padding=3,
bias=False)
self.sn1 = sn.SwitchNorm2d(
64,
using_moving_average=self.using_moving_average,
using_bn=self.using_bn)
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.sn_out = sn.SwitchNorm2d(
512 * block.expansion,
using_moving_average=self.using_moving_average,
using_bn=self.using_bn)
self.avgpool = nn.AvgPool2d(7, stride=1)
self.drouput = nn.Dropout(p=0.5)
self.fc = nn.Linear(512 * block.expansion, num_classes)
for m in self.modules():
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, sn.SwitchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
def _make_layer(self, block, planes, blocks, stride=1):
downsample = None
if stride != 1 or self.inplanes != planes * block.expansion:
downsample = nn.Sequential(
nn.Conv2d(self.inplanes, planes * block.expansion,
kernel_size=1, stride=stride, bias=False),
sn.SwitchNorm2d(planes * block.expansion, using_moving_average=self.using_moving_average,
using_bn=self.using_bn),
)
layers = []
layers.append(block(self.inplanes, planes, stride, downsample, using_moving_average=self.using_moving_average,
using_bn=self.using_bn, last_gamma=self.last_gamma))
self.inplanes = planes * block.expansion
for i in range(1, blocks):
layers.append(block(self.inplanes, planes, using_moving_average=self.using_moving_average,
using_bn=self.using_bn, last_gamma=self.last_gamma))
return nn.Sequential(*layers)
def __init__(self,
layers=50,
bins=(1, 2, 3, 6),
dropout=0.1,
classes=2,
zoom_factor=8,
use_softmax=True,
use_aux=True,
pretrained=False,
syncbn=True):
super(ResNet_Seg_SN, self).__init__()
assert layers in [50, 101, 152]
assert 2048 % len(bins) == 0
assert classes > 1
assert zoom_factor in [1, 2, 4, 8]
self.zoom_factor = zoom_factor
self.use_softmax = use_softmax
self.use_aux = use_aux
# if syncbn:
# from lib.syncbn import SynchronizedBatchNorm2d as BatchNorm
# else:
# from torch.nn import BatchNorm2d as BatchNorm
# models.BatchNorm = BatchNorm
if layers == 50:
resnet_sn = models.resnetv1sn50()
elif layers == 101:
resnet_sn = models.resnetv1sn101()
else:
resnet_sn = models.resnetv1sn152()
self.layer0 = nn.Sequential(resnet_sn.conv1, resnet_sn.sn1,
resnet_sn.relu, resnet_sn.maxpool)
self.layer1, self.layer2, self.layer3, self.layer4 = resnet_sn.layer1, resnet_sn.layer2, resnet_sn.layer3, resnet_sn.layer4
for n, m in self.layer3.named_modules():
if 'conv2' in n:
m.dilation, m.padding, m.stride = (2, 2), (2, 2), (1, 1)
elif 'downsample.0' in n:
m.stride = (1, 1)
for n, m in self.layer4.named_modules():
if 'conv2' in n:
m.dilation, m.padding, m.stride = (4, 4), (4, 4), (1, 1)
elif 'downsample.0' in n:
m.stride = (1, 1)
fea_dim = 2048
self.cls = nn.Sequential(
nn.Conv2d(fea_dim, 512, kernel_size=3, padding=1, bias=False),
#SwitchNorm2d(512, using_bn=True),
sn.SwitchNorm2d(512, using_bn=True),
nn.ReLU(inplace=True),
nn.Dropout2d(p=dropout),
nn.Conv2d(512, classes, kernel_size=1))
if use_aux:
self.aux = nn.Sequential(
nn.Conv2d(1024, 256, kernel_size=3, padding=1, bias=False),
#SwitchNorm2d(256, using_bn=True),
sn.SwitchNorm2d(256, using_bn=True),
nn.ReLU(inplace=True),
nn.Dropout2d(p=dropout),
nn.Conv2d(256, classes, kernel_size=1))