def _resnet(arch, block, layers, pretrained, progress, **kwargs):
model = ResNet(block, layers, **kwargs)
if pretrained:
state_dict = load_state_dict_from_url(model_urls[arch],
progress=progress)
model.load_state_dict(state_dict)
return model
def _resnet(arch: str, in_channels, block: Type[Union[BasicBlock, Bottleneck]],
layers: List[int], layers_channel: List[int], pretrained: bool,
progress: bool, **kwargs: Any) -> ResNetM:
model = ResNetM(block, in_channels, layers, layers_channel, **kwargs)
if pretrained:
state_dict = load_state_dict_from_url(model_urls[arch],
progress=progress)
model.load_state_dict(state_dict)
return model
def _resnet_global_local_pred(codeword_multiplier, out_dims,
arch, block, layers, pretrained, progress, **kwargs):
model = GlobalLocalPredResNet(codeword_multiplier, out_dims,
block, layers, **kwargs)
if pretrained:
state_dict = ResNetUtils.load_state_dict_from_url(ResNetUtils.model_urls[arch],
progress=progress)
model.load_state_dict(state_dict)
return model
def resnet152(pretrained=True, is_lower=True, num_classes=12, **kwargs):
model = ResNetLower(Bottleneck, [3, 8, 36, 2], **
kwargs) if is_lower else ResNetUpper(
num_classes=num_classes)
if pretrained:
model.load_imagenet_state_dict(
resnet.load_state_dict_from_url(model_urls['resnet152']))
print("loaded imagenet pretrained resnet152")
return model
def ResNet18(pretrained=False, fine_tune=False, classes=(10, 10), pool=True):
model = MultiTaskResNet(fine_tune, classes, pool)
if pretrained:
state_dict = resnet.load_state_dict_from_url(
resnet.model_urls['resnet18'])
model.load_state_dict(state_dict)
return model
def _resnetlstm(arch, pretrained, block, layers, **kwargs):
model = ResNetLSTM(**kwargs)
# use weights pretrained on the 1000-class ImageNet dataset
if pretrained:
state_dict = load_state_dict_from_url(model_urls[arch])
for key in model.state_dict():
if 'lstm' in key or 'fc' in key:
state_dict[key] = model.state_dict()[key]
model.load_state_dict(state_dict)
return model
def _resnet(codeword_multiplier, sparsity_multiplier, arch, block, layers,
pretrained, progress, **kwargs):
model = SparseRepResNet(codeword_multiplier, sparsity_multiplier, block,
layers, **kwargs)
if pretrained:
state_dict = ResNetUtils.load_state_dict_from_url(
ResNetUtils.model_urls[arch], progress=progress)
model.load_state_dict(state_dict)
return model
def load_pretrained_resnet(arch, progress=True):
if arch == 'resnet34':
model = ResNet(BasicBlock, [3, 4, 6, 3])
if arch == 'resnet18':
model = ResNet(BasicBlock, [2, 2, 2, 2])
state_dict = load_state_dict_from_url(model_urls[arch], progress=progress)
model.load_state_dict(state_dict, strict=False)
model.fc_channel = model.fc.in_features
delattr(model, 'fc')
return model
def _resnet(arch, block, layers, branches, pretrained, progress, **kwargs):
model = MTLResNet(block, layers, branches, **kwargs)
if pretrained:
state_dict = load_state_dict_from_url(model_urls[arch],
progress=progress)
own_state = model.state_dict()
pretrained_dict = {
k: v
for k, v in state_dict.items() if k in own_state
}
# 2. overwrite entries in the existing state dict
own_state.update(pretrained_dict)
# 3. load the new state dict
model.load_state_dict(own_state)
return model
def __init__(self,
landmark_num: int = 5,
pretrained=True,
pretrained_path: str = None):
super().__init__()
self.res = torchvision.models.resnet18(pretrained=False,
num_classes=landmark_num * 2)
if pretrained:
if pretrained_path:
state_dict = torch.load(pretrained_path)
self.load_state_dict(state_dict, strict=True)
else:
state_dict = load_state_dict_from_url(
res_model_urls['resnet18'], progress=True)
state_dict.pop('fc.weight')
state_dict.pop('fc.bias')
self.res.load_state_dict(state_dict, strict=False)
