def resnet18nodownsample(pretrained=False, **kwargs):
"""Constructs a ResNet-18 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(BasicBlock, [2, 2, 2, 2], nodownsampling=True, **kwargs)
if pretrained:
model.load_state_dict(model_zoo.load_url(model_urls['resnet18']))
model.name = 'ResNet18NoDownsampling'
return model
def resnet18_ep0(pretrained=False, **kwargs):
"""Constructs a ResNet-18 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(BasicBlock, [2, 2, 2, 2], disable_early_pooling=True, **kwargs)
if pretrained:
model.load_state_dict(model_zoo.load_url(model_urls['resnet18']))
model.name = 'ResNet18_EP0'
return model
def resnet18_dp00_dspl2(pretrained=False, **kwargs):
"""Constructs a ResNet-18 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(BasicBlock, [4, 4, None, None], disable_early_pooling=True, disable_early_downsampling=True, **kwargs)
if pretrained:
model.load_state_dict(model_zoo.load_url(model_urls['resnet18']))
model.name = 'ResNet18_dp00_DSPL2'
return model
def se_resnet50(num_classes=1000, pretrained=False):
"""Constructs a ResNet-50 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(SEBottleneck, [3, 4, 6, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
if pretrained:
model.load_state_dict(model_zoo.load_url("https://www.dropbox.com/s/xpq8ne7rwa4kg4c/seresnet50-60a8950a85b2b.pkl"))
return model
def resnet152(pretrained=False, **kwargs):
"""Constructs a ResNet-152 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(Bottleneck, [3, 8, 36, 3], **kwargs)
if pretrained:
model.load_state_dict(model_zoo.load_url(model_urls['resnet152']))
model.name = 'ResNet152'
return model
def resnet18_dspl5(pretrained=False, **kwargs):
"""Constructs a ResNet-18 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(BasicBlock, [1, 1, 2, 2, 2, None, None, None], **kwargs)
if pretrained:
model.load_state_dict(model_zoo.load_url(model_urls['resnet18']))
model.name = 'ResNet18_DSPL5'
return model
def se_resnet50(num_classes=1000, pretrained=False):
"""Constructs a ResNet-50 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(SEBottleneck, [3, 4, 6, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
if pretrained:
model.load_state_dict(load_state_dict_from_url(
"https://github.com/moskomule/senet.pytorch/releases/download/archive/seresnet50-60a8950a85b2b.pkl"))
return model
def resnet18_XXS(pretrained=False, **kwargs):
"""Constructs a ResNet-18 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(BasicBlock, [2, 2, 2, 2], scale_factor=8, **kwargs)
if pretrained:
model.load_state_dict(model_zoo.load_url(model_urls['resnet18']))
model.name = 'ResNet18_XXS'
return model
def resnet18noskip_dspl3(pretrained=False, **kwargs):
"""Constructs a ResNet-18 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
kwargs["noskip"] = True
model = ResNet(BasicBlock, [2, 3, 3, None], **kwargs)
if pretrained:
model.load_state_dict(model_zoo.load_url(model_urls['resnet18']))
model.name = 'ResNet18NoSkip_DSPL3'
return model
def resnet34(pretrained=False, **kwargs):
"""Constructs a ResNet-34 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(BasicBlock, [3, 4, 6, 3], **kwargs)
if pretrained:
model.load_state_dict(model_zoo.load_url(model_urls['resnet34']))
model.name = 'ResNet34'
return model
def resnet18noskip(pretrained=False, **kwargs):
"""Constructs a ResNet-18 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
if "noskip" in kwargs:
kwargs.pop("noskip")
model = ResNet(BasicBlock, [2, 2, 2, 2], noskip=True, **kwargs)
if pretrained:
model.load_state_dict(model_zoo.load_url(model_urls['resnet18']))
model.name = 'ResNet18NoSkip'
return model
def resnet50noskip(pretrained=False, **kwargs):
"""Constructs a ResNet-50 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
if "noskip" in kwargs:
kwargs.pop("noskip")
model = ResNet(Bottleneck, [3, 4, 6, 3], noskip=True, **kwargs)
if pretrained:
model.load_state_dict(model_zoo.load_url(model_urls['resnet50']))
model.name = 'ResNet50NoSkip'
return model
def resnet152(pretrained=False, **kwargs):
if pretrained:
model = ResNet(Bottleneck, [3, 8, 36, 3], **kwargs)
pretrained_state_dict = torch.load(
'./Authority/resnet152-b121ed2d.pth'
) # load_url函数根据model_urls字典下载或导入相应的预训练模型
now_state_dict = model.state_dict() # 返回model模块的字典
pretrained_state_dict.pop('fc.weight') # 排除全连接层的参数(全连接层返回分类个数)
pretrained_state_dict.pop('fc.bias')
now_state_dict.update(pretrained_state_dict)
model.load_state_dict(now_state_dict)
# 最后通过调用model的load_state_dict方法用预训练的模型参数来初始化你构建的网络结构，
# 这个方法就是PyTorch中通用的用一个模型的参数初始化另一个模型的层的操作。load_state_dict方法还有一个重要的参数是strict，
# 该参数默认是True，表示预训练模型的层和你的网络结构层严格对应相等（比如层名和维度）
return model
return ResNet(Bottleneck, [3, 8, 36, 3], **kwargs)
def se_resnet34(num_classes=1000, pretrained="/home/ibian/.torch/models/resnet34-333f7ec4.pth"):
"""Constructs a ResNet-34 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(SEBasicBlock, [3, 4, 6, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
# use pretrained model
if pretrained is not None:
print("=> loading pretrained model '{}'".format(pretrained))
model_dict = model.state_dict()
checkpoint = torch.load(pretrained)
checkpoint = {k: v for k, v in checkpoint.items() if k in model_dict}
model_dict.update(checkpoint)
model.load_state_dict(model_dict)
return model
def resnet18_domain(classes=4, domains=30, url=None, **kwargs):
"""Constructs a ResNet-18 model.
Args:
num_classes (int): the number of classes of the classification model
domains (int): the number of domains included in the model (#source + #auxuliary)
"""
# Instantiate original ResNet
model_origin = ResNet(BasicBlock, [2, 2, 2, 2], **kwargs)
model_origin.load_state_dict(model_zoo.load_url(url))
# Instatiate Domain ResNet
model = DomainResNet(DomainBlock, [2, 2, 2, 2], domains=domains, **kwargs)
# Copy BN stats and params from the original to the domain-based
c = 0
for m_orig in model_origin.named_modules():
if 'bn' in m_orig[0] or 'downsample.1' in m_orig[0]:
for m_doms in model.named_modules():
if m_doms[0] == m_orig[0]:
c += 1
for i in range(domains):
m_doms[1].bns[i].running_var.data[:] = m_orig[
1].running_var.data[:]
m_doms[1].bns[i].running_mean.data[:] = m_orig[
1].running_mean.data[:]
m_doms[1].scale.data[i, :] = m_orig[1].weight.data[:]
m_doms[1].bias.data[i, :] = m_orig[1].bias.data[:]
elif 'conv' in m_orig[0] or 'downsample.0' in m_orig[0]:
for m_doms in model.named_modules():
if m_doms[0] == m_orig[0]:
m_doms[1].weight.data[:] = m_orig[1].weight.data[:]
# Init classifier
num_ftrs = model.fc.in_features
model.fc = nn.Linear(num_ftrs, classes)
for m in model.modules():
if isinstance(m, nn.Linear):
m.weight.data.normal_(0, 0.0001)
return model
class ClippedResNet(nn.Module):
"""
"""
def __init__(self, pretrained=False, out_features=2):
"""
"""
super().__init__()
self.net = ResNet(BasicBlock, [2, 2, 2, 2])
if pretrained:
self.net.load_state_dict(
model_zoo.load_url(model_urls['resnet_18']))
# change last layer
self.net.fc = nn.Linear(self.net.fc.in_features, out_features)
self.out_features = out_features
def forward(self, input):
"""
"""
return self.net(input)
model = ResNet(SEBasicBlock, [3, 4, 6, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
return model
def se_resnet50(num_classes=1_000, pretrained=False):
"""Constructs a ResNet-50 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(SEBottleneck, [3, 4, 6, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
if pretrained:
model.load_state_dict(
load_state_dict_from_url(
"https://github.com/moskomule/senet.pytorch/releases/download/archive/seresnet50-60a8950a85b2b.pkl"
))
return model
def se_resnet101(num_classes=1_000):
"""Constructs a ResNet-101 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(SEBottleneck, [3, 4, 23, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
return model
"""
model = ResNet(SEBasicBlock, [3, 4, 6, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
return model
def se_resnet50(num_classes=1_000, pretrained=False):
"""Constructs a ResNet-50 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(SEBottleneck, [3, 4, 6, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
if pretrained:
model.load_state_dict(model_zoo.load_url("https://www.dropbox.com/s/xpq8ne7rwa4kg4c/seresnet50-60a8950a85b2b.pkl"))
return model
def se_resnet101(num_classes=1_000):
"""Constructs a ResNet-101 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(SEBottleneck, [3, 4, 23, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
return model
def se_resnet152(num_classes=1_000):
class UNetResNet(nn.Module):
"""PyTorch U-Net model using ResNet(34, 50, 101 or 152) encoder.
UNet: https://arxiv.org/abs/1505.04597
ResNet: https://arxiv.org/abs/1512.03385
Proposed by Alexander Buslaev: https://www.linkedin.com/in/al-buslaev/
Args:
encoder_depth (int): Depth of a ResNet encoder (34, 101 or 152).
num_classes (int): Number of output classes.
num_filters (int, optional): Number of filters in the last layer of decoder. Defaults to 32.
dropout_2d (float, optional): Probability factor of dropout layer before output layer. Defaults to 0.2.
pretrained (bool, optional):
False - no pre-trained weights are being used.
True - ResNet encoder is pre-trained on ImageNet.
Defaults to False.
is_deconv (bool, optional):
False: bilinear interpolation is used in decoder.
True: deconvolution is used in decoder.
Defaults to False.
"""
def __init__(self,
encoder_depth,
num_classes,
num_filters=32,
dropout_2d=0.2,
pretrained=False,
is_deconv=False):
super().__init__()
self.num_classes = num_classes
self.dropout_2d = dropout_2d
if encoder_depth == 34:
self.encoder = ResNet(SEBasicBlock, [3, 4, 6, 3])
if pretrained:
self.encoder.load_state_dict(model_zoo.load_url(
model_urls["resnet34"]),
strict=False)
bottom_channel_nr = 512
elif encoder_depth == 50:
self.encoder = ResNet(SEBottleneck, [3, 4, 6, 3])
if pretrained:
self.encoder.load_state_dict(model_zoo.load_url(
model_urls["resnet50"]),
strict=False)
bottom_channel_nr = 2048
elif encoder_depth == 101:
self.encoder = ResNet(SEBottleneck, [3, 4, 23, 3])
if pretrained:
self.encoder.load_state_dict(model_zoo.load_url(
model_urls["resnet101"]),
strict=False)
bottom_channel_nr = 2048
elif encoder_depth == 152:
# self.encoder = torchvision.models.resnet152(pretrained=pretrained)
self.encoder = ResNet(SEBottleneck, [3, 8, 36, 3])
if pretrained:
self.encoder.load_state_dict(model_zoo.load_url(
model_urls["resnet152"]),
strict=False)
bottom_channel_nr = 2048
else:
raise NotImplementedError(
'only 34, 50, 101, 152 version of Resnet are implemented')
self.input_adjust = nn.Sequential(self.encoder.conv1, self.encoder.bn1,
self.encoder.relu)
self.conv1 = self.encoder.layer1
self.conv2 = self.encoder.layer2
self.conv3 = self.encoder.layer3
self.conv4 = self.encoder.layer4
self.dec4 = DecoderBlockV2(bottom_channel_nr, num_filters * 8 * 2,
num_filters * 8, is_deconv)
self.dec3 = DecoderBlockV2(bottom_channel_nr // 2 + num_filters * 8,
num_filters * 8 * 2, num_filters * 8,
is_deconv)
self.dec2 = DecoderBlockV2(bottom_channel_nr // 4 + num_filters * 8,
num_filters * 4 * 2, num_filters * 2,
is_deconv)
self.dec1 = DecoderBlockV2(bottom_channel_nr // 8 + num_filters * 2,
num_filters * 2 * 2, num_filters * 2 * 2,
is_deconv)
self.final = nn.Conv2d(num_filters * 2 * 2, num_classes, kernel_size=1)
def forward(self, x):
input_adjust = self.input_adjust(x)
conv1 = self.conv1(input_adjust)
conv2 = self.conv2(conv1)
conv3 = self.conv3(conv2)
center = self.conv4(conv3)
dec4 = self.dec4(center)
dec3 = self.dec3(torch.cat([dec4, conv3], 1))
dec2 = self.dec2(torch.cat([dec3, conv2], 1))
dec1 = F.dropout2d(self.dec1(torch.cat([dec2, conv1], 1)),
p=self.dropout_2d)
return self.final(dec1)
model = ResNet(SEBasicBlock, [3, 4, 6, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
return model
def se_resnet50(num_classes=1_000, pretrained=False):
"""Constructs a ResNet-50 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(SEBottleneck, [3, 4, 6, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
if pretrained:
model.load_state_dict(
model_zoo.load_url(
"https://www.dropbox.com/s/xpq8ne7rwa4kg4c/seresnet50-f6e7c9de8bf52b5.pkl"
))
return model
def se_resnet101(num_classes=1_000):
"""Constructs a ResNet-101 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(SEBottleneck, [3, 4, 23, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
return model
return model
def se_resnet50(num_classes=1_000, pretrained=False):
"""Constructs a ResNet-50 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(SEBottleneck, [3, 4, 6, 3])
model.avgpool = nn.AdaptiveAvgPool2d(1)
if pretrained:
# state_dict = load_state_dict_from_url(model_urls['resnet50'],
# progress=progress)
# model.load_state_dict(state_dict)
model.load_state_dict(torch.load(PATH))
num_features = model.fc.in_features
model.fc = nn.Linear(num_features, num_classes)
return model
def se_resnet101(num_classes=1_000):
"""Constructs a ResNet-101 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(SEBottleneck, [3, 4, 23, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
return model
# model.avgpool = nn.AdaptiveAvgPool2d(1)
# return model
#
def se_resnet50(num_classes=1_000, pretrained=False):
"""Constructs a ResNet-50 model.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
model = ResNet(SEBottleneck, [3, 4, 6, 3], num_classes=num_classes)
model.avgpool = nn.AdaptiveAvgPool2d(1)
if pretrained:
model.load_state_dict(
model_zoo.load_url(
"http://data.lip6.fr/cadene/pretrainedmodels/se_resnet50-ce0d4300.pth"
))
return model
# def se_resnet101(num_classes=1_000):
# """Constructs a ResNet-101 model.
#
# Args:
# pretrained (bool): If True, returns a model pre-trained on ImageNet
# """
# model = ResNet(SEBottleneck, [3, 4, 23, 3], num_classes=num_classes)
# model.avgpool = nn.AdaptiveAvgPool2d(1)
# return model
#
#