def __init__(self, inplanes, planes, stride=1, downsample=None):
super(BasicBlock, self).__init__()
self.conv1 = conv3x3(inplanes, planes, stride)
self.bn1 = SynchronizedBatchNorm2d(planes)
self.relu = nn.ReLU(inplace=True)
self.conv2 = conv3x3(planes, planes)
self.bn2 = SynchronizedBatchNorm2d(planes)
self.downsample = downsample
self.stride = stride
def __init__(self, inplanes, planes, stride=1, downsample=None):
super(Bottleneck, self).__init__()
self.conv1 = conv1x1(inplanes, planes)
self.bn1 = SynchronizedBatchNorm2d(planes)
self.conv2 = conv3x3(planes, planes, stride)
self.bn2 = SynchronizedBatchNorm2d(planes)
self.conv3 = conv1x1(planes, planes * self.expansion)
self.bn3 = SynchronizedBatchNorm2d(planes * self.expansion)
self.relu = nn.ReLU(inplace=True)
self.downsample = downsample
self.stride = stride
def __init__(self, in_channels, out_channels, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), dilation=1):
super(ConvBn2d, self).__init__()
self.conv = nn.Conv2d(in_channels, out_channels, kernel_size=kernel_size, stride=stride, padding=padding,
bias=False, dilation=dilation)
#self.bn = nn.BatchNorm2d(out_channels)
self.bn = SynchronizedBatchNorm2d(out_channels)
def __init__(self, in_channel):
super(RefineBlock, self).__init__()
self.c1 = nn.Conv2d(in_channel, 512, kernel_size=1, stride=1, padding=0, bias=False)
self.c3_1 = nn.Conv2d(512, 512, kernel_size=3, stride=1, padding=1, bias=False)
self.bn = SynchronizedBatchNorm2d(512)
self.relu = nn.ReLU(inplace=True)
self.c3_2 = nn.Conv2d(512, 512, kernel_size=3, stride=1, padding=1, bias=False)
def __init__(self, in_size, out_size):
super(GAU, self).__init__()
self.in_size = in_size
self.out_size = out_size
self.conv = nn.Conv2d(in_size*2, out_size, kernel_size=1, stride=1, bias=False)
self.avg_pool = nn.AdaptiveAvgPool2d(1)
self.bn=SynchronizedBatchNorm2d(in_size)
self.relu=nn.ReLU(inplace=True)
def basic_bn_shortcut(inplanes, outplanes, stride):
return nn.Sequential(
nn.Conv2d(inplanes,
outplanes,
kernel_size=1,
stride=stride,
bias=False),
SynchronizedBatchNorm2d(outplanes),
)
def __init__(self,
inplanes,
outplanes,
innerplanes,
stride=1,
dilation=1,
group=1,
downsample=None):
super().__init__()
# In original resnet, stride=2 is on 1x1.
# In fb.torch resnet, stride=2 is on 3x3.
(str1x1, str3x3) = (stride, 1) if cfg.RESNETS.STRIDE_1X1 else (1,
stride)
self.stride = stride
self.conv1 = nn.Conv2d(inplanes,
innerplanes,
kernel_size=1,
stride=str1x1,
bias=False)
self.bn1 = SynchronizedBatchNorm2d(innerplanes)
self.conv2 = nn.Conv2d(innerplanes,
innerplanes,
kernel_size=3,
stride=str3x3,
bias=False,
padding=1 * dilation,
dilation=dilation,
groups=group)
self.bn2 = SynchronizedBatchNorm2d(innerplanes)
self.conv3 = nn.Conv2d(innerplanes,
outplanes,
kernel_size=1,
stride=1,
bias=False)
self.bn3 = SynchronizedBatchNorm2d(outplanes)
self.downsample = downsample
self.relu = nn.ReLU(inplace=True)
def _make_layer(self, block, planes, blocks, stride=1):
downsample = None
if stride != 1 or self.inplanes != planes * block.expansion:
downsample = nn.Sequential(
conv1x1(self.inplanes, planes * block.expansion, stride),
SynchronizedBatchNorm2d(planes * block.expansion),
)
layers = []
layers.append(block(self.inplanes, planes, stride, downsample))
self.inplanes = planes * block.expansion
for _ in range(1, blocks):
layers.append(block(self.inplanes, planes))
return nn.Sequential(*layers)
def add_norm_layer(layer):
if norm_type == 'batch':
norm_layer = nn.BatchNorm2d(get_out_channel(layer),
affine=True,
track_running_stats=True)
elif norm_type == 'sync_batch':
norm_layer = SynchronizedBatchNorm2d(get_out_channel(layer),
affine=True)
elif norm_type == 'instance':
norm_layer = nn.InstanceNorm2d(get_out_channel(layer),
affine=False)
else:
raise ValueError('normalization layer %s is not recognized' %
norm_type)
return norm_layer
def __init__(self,
in_channels,
out_channels,
kernel,
stride=1,
activation=None,
use_bn=False):
super(BaseConv, self).__init__()
self.use_bn = use_bn
self.activation = activation
self.conv = nn.Conv2d(in_channels, out_channels, kernel, stride,
kernel // 2)
self.conv.weight.data.normal_(0, 0.01)
self.conv.bias.data.zero_()
self.bn = SynchronizedBatchNorm2d(out_channels)
self.bn.weight.data.fill_(1)
self.bn.bias.data.zero_()
def __init__(self, in_channel, out_channel):
super(FPA, self).__init__()
self.c15_1 = nn.Conv2d(in_channel, out_channel, kernel_size=15, stride=1, padding=7, bias=False)
self.c11_1 = nn.Conv2d(in_channel, out_channel, kernel_size=11, stride=1, padding=5, bias=False)
self.c7_1 = nn.Conv2d(in_channel, out_channel, kernel_size=7, stride=1, padding=3, bias=False)
self.c3_1 = nn.Conv2d(in_channel, out_channel, kernel_size=3, stride=1, padding=1, bias=False)
self.c15_2 = nn.Conv2d(in_channel, out_channel, kernel_size=15, stride=1, padding=7, bias=False)
self.c11_2 = nn.Conv2d(in_channel, out_channel, kernel_size=11, stride=1, padding=5, bias=False)
self.c7_2 = nn.Conv2d(in_channel, out_channel, kernel_size=7, stride=1, padding=3, bias=False)
self.c3_2 = nn.Conv2d(in_channel, out_channel, kernel_size=3, stride=1, padding=1, bias=False)
self.avg_pool = nn.AdaptiveAvgPool2d(1)
self.c1_gpb = nn.Conv2d(in_channel, out_channel, kernel_size=1, bias=False)
self.bn = SynchronizedBatchNorm2d(out_channel)
self.relu = nn.ReLU(inplace=True)
def __init__(self,
in_channels,
out_channels,
kernel_size=3,
padding=1,
dilation=1,
stride=1,
groups=1,
is_bn=True):
super(ConvBn2d, self).__init__()
self.conv = nn.Conv2d(in_channels,
out_channels,
kernel_size=kernel_size,
padding=padding,
stride=stride,
dilation=dilation,
groups=groups,
bias=False)
self.bn = SynchronizedBatchNorm2d(out_channels)
def __init__(self, block, layers, num_classes=1000):
super(MV3_1_true_2_ResNet, self).__init__()
# self.do = nn.Dropout(p=0.5)
self.inplanes = 64
self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,
bias=False)
self.bn1 = SynchronizedBatchNorm2d(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.rb1_1 = RefineBlock(256)
self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
self.rb2_1 = RefineBlock(512)
self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
self.rb3_1 = RefineBlock(1024)
self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
self.rb4_1 = RefineBlock(2048)
# self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
# self.fc = nn.Linear(512 * block.expansion, num_classes)
# only for >=res50
# self.fpa=FPA(2048,512)
self.fpa = FPA(512, 512)
self.rb4_2 = RefineBlock(512 * 5)
self.fuse43 = GAU(512, 512)
# self.post_proc43 = conv3x3_bn(512*2,512)
self.rb3_2 = RefineBlock(512)
self.fuse32 = GAU(512, 512)
self.rb2_2 = RefineBlock(512)
# self.post_proc32 = conv3x3_bn(512)
self.fuse21 = GAU(512, 512)
self.rb1_2 = RefineBlock(512)
# self.post_proc21 = conv3x3_bn(512)
self.class_conv = nn.Conv2d(512, num_classes, kernel_size=3, stride=1,
padding=1, bias=True)
def sync_batchnorm_2d(in_features, eps=1e-5, momentum=0.0001, affine=True):
return SynchronizedBatchNorm2d(in_features,
eps=eps,
momentum=momentum,
affine=affine)