| """ |
| Backbones supported by torchvison. |
| """ |
| import torch |
| import torch.nn as nn |
| import torchvision |
|
|
|
|
| class Res101Encoder(nn.Module): |
| """ |
| Resnet101 backbone from deeplabv3 |
| modify the 'downsample' component in layer2 and/or layer3 and/or layer4 as the vanilla Resnet |
| """ |
|
|
| def __init__(self, replace_stride_with_dilation=None, pretrained_weights='resnet101'): |
| super().__init__() |
| |
| if pretrained_weights == 'deeplabv3': |
| self.pretrained_weights = torch.load( |
| "./deeplabv3_resnet101_coco-586e9e4e.pth", map_location='cpu') |
| elif pretrained_weights == 'resnet101': |
| self.pretrained_weights = torch.load("./model1/resnet101-63fe2227.pth", |
| map_location='cpu') |
| else: |
| self.pretrained_weights = pretrained_weights |
|
|
| _model = torchvision.models.resnet.resnet101(pretrained=False, |
| replace_stride_with_dilation=replace_stride_with_dilation) |
| self.backbone = nn.ModuleDict() |
| for dic, m in _model.named_children(): |
| self.backbone[dic] = m |
|
|
| self.reduce1 = nn.Conv2d(1024, 512, kernel_size=1, bias=False) |
| self.reduce2 = nn.Conv2d(2048, 512, kernel_size=1, bias=False) |
| self.reduce1d = nn.Linear(in_features=1000, out_features=1, bias=True) |
|
|
| self._init_weights() |
|
|
| def forward(self, x): |
| features = dict() |
| x = self.backbone["conv1"](x) |
| x = self.backbone["bn1"](x) |
| x = self.backbone["relu"](x) |
| |
| x = self.backbone["maxpool"](x) |
| x = self.backbone["layer1"](x) |
| x = self.backbone["layer2"](x) |
| x = self.backbone["layer3"](x) |
| features['down2'] = self.reduce1(x) |
| x = self.backbone["layer4"](x) |
| features['down3'] = self.reduce2(x) |
| |
| |
| t = self.backbone["avgpool"](x) |
| t = torch.flatten(t, 1) |
| t = self.backbone["fc"](t) |
| t = self.reduce1d(t) |
| return (features, t) |
|
|
| def _init_weights(self): |
| for m in self.modules(): |
| if isinstance(m, nn.Conv2d): |
| nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu') |
| elif isinstance(m, nn.BatchNorm2d): |
| nn.init.constant_(m.weight, 1) |
| nn.init.constant_(m.bias, 0) |
|
|
| if self.pretrained_weights is not None: |
| keys = list(self.pretrained_weights.keys()) |
| new_dic = self.state_dict() |
| new_keys = list(new_dic.keys()) |
|
|
| for i in range(len(keys)): |
| if keys[i] in new_keys: |
| new_dic[keys[i]] = self.pretrained_weights[keys[i]] |
|
|
| self.load_state_dict(new_dic) |
|
|
