def __init__(self,
in_channels,
channels,
stride,
downsample=False,
last_gamma=False,
**kwargs):
super(BasicBlockV1, self).__init__(**kwargs)
self.body = list()
self.body.append(_conv3x3(in_channels, channels, stride))
self.body.append(nn.BatchNorm2d(channels))
self.body.append(nn.ReLU(inplace=True))
self.body.append(_conv3x3(channels, channels, 1))
tmp_layer = nn.BatchNorm2d(channels)
if last_gamma:
nn.init.zeros_(tmp_layer.weight)
self.body.append(tmp_layer)
self.body = nn.Sequential(*self.body)
if downsample:
self.downsample = nn.Sequential(
nn.Conv2d(in_channels,
channels,
kernel_size=1,
stride=stride,
bias=False), nn.BatchNorm2d(channels))
else:
self.downsample = None
def __init__(self, in_channels, channels, stride, downsample=False, **kwargs):
super(CIFARBasicBlockV2, self).__init__(**kwargs)
self.bn1 = nn.BatchNorm2d(in_channels)
self.conv1 = _conv3x3(in_channels, channels, stride)
self.bn2 = nn.BatchNorm2d(channels)
self.conv2 = _conv3x3(channels, channels, 1)
if downsample:
self.downsample = nn.Conv2d(in_channels, channels, 1, stride, bias=False)
else:
self.downsample = None
def __init__(self, in_channels, channels, stride, downsample=False, **kwargs):
super(CIFARBasicBlockV1, self).__init__(**kwargs)
self.body = nn.Sequential(_conv3x3(in_channels, channels, stride),
nn.BatchNorm2d(channels), nn.ReLU(inplace=True),
_conv3x3(channels, channels, 1),
nn.BatchNorm2d(channels))
if downsample:
self.downsample = nn.Sequential(nn.Conv2d(in_channels, channels, 1, stride=stride, bias=False),
nn.BatchNorm2d(channels))
else:
self.downsample = None
def __init__(self,
block,
layers,
channels,
classes=1000,
thumbnail=False,
last_gamma=False,
**kwargs):
super(ResNetV1, self).__init__(**kwargs)
assert len(layers) == len(channels) - 2
self.features = list()
if thumbnail:
self.features.append(_conv3x3(channels[0], channels[1], 1))
else:
self.features.append(
nn.Conv2d(channels[0], channels[1], 7, 2, 3, bias=False))
self.features.append(nn.BatchNorm2d(channels[1]))
self.features.append(nn.ReLU(inplace=True))
self.features.append(nn.MaxPool2d(3, 2, 1))
for i, num_layer in enumerate(layers):
stride = 1 if i == 0 else 2
self.features.append(
self._make_layer(block,
num_layer,
channels[i + 1],
channels[i + 2],
stride,
last_gamma=last_gamma))
self.features = nn.Sequential(*self.features)
self.output = nn.Linear(channels[-1], classes)
def __init__(self,
in_channels,
channels,
stride,
downsample=False,
last_gamma=False,
**kwargs):
super(BottleneckV2, self).__init__(**kwargs)
self.bn1 = nn.BatchNorm2d(in_channels)
self.conv1 = nn.Conv2d(in_channels,
channels // 4,
kernel_size=1,
stride=1,
bias=False)
self.bn2 = nn.BatchNorm2d(channels // 4)
self.conv2 = _conv3x3(channels // 4, channels // 4, stride)
self.bn3 = nn.BatchNorm2d(channels // 4)
if last_gamma:
nn.init.zeros_(self.bn3.weight)
self.conv3 = nn.Conv2d(channels // 4,
channels,
kernel_size=1,
stride=1,
bias=False)
if downsample:
self.downsample = nn.Conv2d(in_channels,
channels,
1,
stride,
bias=False)
else:
self.downsample = None
def __init__(self,
in_channels,
channels,
stride,
downsample=False,
last_gamma=False,
**kwargs):
super(BasicBlockV2, self).__init__(**kwargs)
self.bn1 = nn.BatchNorm2d(in_channels)
self.conv1 = _conv3x3(in_channels, channels, stride)
self.bn2 = nn.BatchNorm2d(channels)
if last_gamma:
nn.init.zeros_(self.bn2.weight)
self.conv2 = _conv3x3(channels, channels, 1)
if downsample:
self.downsample = nn.Conv2d(in_channels,
channels,
1,
stride,
bias=False)
else:
self.downsample = None
