def __init__(self, num_classes, layer='50', input_ch=3):
super(zzResBase, self).__init__()
self.num_classes = num_classes
print('resnet' + layer)
if layer == '18':
resnet = extended_resnet.resnet18(pretrained=True,
input_ch=input_ch)
elif layer == '50':
resnet = extended_resnet.resnet50(pretrained=True,
input_ch=input_ch)
elif layer == '101':
resnet = extended_resnet.resnet101(pretrained=True,
input_ch=input_ch)
elif layer == '152':
resnet = extended_resnet.resnet152(pretrained=True,
input_ch=input_ch)
else:
NotImplementedError
self.conv1 = resnet.conv1
self.bn0 = resnet.bn1
self.relu = resnet.relu
self.maxpool = resnet.maxpool
self.layer1 = resnet.layer1
self.layer2 = resnet.layer2
self.layer3 = resnet.layer3
self.layer4 = resnet.layer4
def __init__(self, num_classes, layer='50', input_ch=3):
super(ResFCN, self).__init__()
self.num_classes = num_classes
print('resnet' + layer)
if layer == '18':
resnet = extended_resnet.resnet18(pretrained=True,
input_ch=input_ch)
elif layer == '34':
resnet = extended_resnet.resnet34(pretrained=True,
input_ch=input_ch)
elif layer == '50':
resnet = extended_resnet.resnet50(pretrained=True,
input_ch=input_ch)
elif layer == '101':
resnet = extended_resnet.resnet101(pretrained=True,
input_ch=input_ch)
elif layer == '152':
resnet = extended_resnet.resnet152(pretrained=True,
input_ch=input_ch)
else:
NotImplementedError
self.conv1 = resnet.conv1
self.bn0 = resnet.bn1
self.relu = resnet.relu
self.maxpool = resnet.maxpool
self.layer1 = resnet.layer1
self.layer2 = resnet.layer2
self.layer3 = resnet.layer3
self.layer4 = resnet.layer4
self.num_classes = num_classes
self.upsample1 = Upsample(2048, 1024)
self.upsample2 = Upsample(1024, 512)
self.upsample3 = Upsample(512, 64)
self.upsample4 = Upsample(64, 64)
self.upsample5 = Upsample(64, 32)
self.fs1 = Fusion(1024)
self.fs2 = Fusion(512)
self.fs3 = Fusion(256)
self.fs4 = Fusion(64)
self.fs5 = Fusion(64)
self.out5 = self._classifier(32)
self.transformer = nn.Conv2d(256, 64, kernel_size=1)
def __init__(self,
base_model='resnet50',
input_ch=3,
use_dropout_at_layer4=True):
super(ResBase, self).__init__()
print(base_model)
if base_model == 'resnet18':
resnet = extended_resnet.resnet18(
pretrained=True,
input_ch=input_ch,
use_dropout_at_layer4=use_dropout_at_layer4)
elif base_model == 'resnet50':
resnet = extended_resnet.resnet50(
pretrained=True,
input_ch=input_ch,
use_dropout_at_layer4=use_dropout_at_layer4)
elif base_model == 'resnet101':
resnet = extended_resnet.resnet101(
pretrained=True,
input_ch=input_ch,
use_dropout_at_layer4=use_dropout_at_layer4)
elif base_model == 'resnet152':
resnet = extended_resnet.resnet152(
pretrained=True,
input_ch=input_ch,
use_dropout_at_layer4=use_dropout_at_layer4)
else:
raise ValueError("{} is not supported".format(base_model))
self.conv1 = resnet.conv1
self.bn0 = resnet.bn1
self.relu = resnet.relu
self.maxpool = resnet.maxpool
self.layer1 = resnet.layer1
self.layer2 = resnet.layer2
self.layer3 = resnet.layer3
self.layer4 = resnet.layer4