| 62 | return x |
| 63 | |
| 64 | class DepthNet(nn.Module): |
| 65 | __factory = { |
| 66 | 18: Resnet.resnet18, |
| 67 | 34: Resnet.resnet34, |
| 68 | 50: Resnet.resnet50, |
| 69 | 101: Resnet.resnet101, |
| 70 | 152: Resnet.resnet152 |
| 71 | } |
| 72 | def __init__(self, |
| 73 | backbone='resnet', |
| 74 | depth=50, |
| 75 | upfactors=[2, 2, 2, 2]): |
| 76 | super(DepthNet, self).__init__() |
| 77 | self.backbone = backbone |
| 78 | self.depth = depth |
| 79 | self.pretrained = False |
| 80 | self.inchannels = [256, 512, 1024, 2048] |
| 81 | self.midchannels = [256, 256, 256, 512] |
| 82 | self.upfactors = upfactors |
| 83 | self.outchannels = 1 |
| 84 | |
| 85 | # Build model |
| 86 | if self.backbone == 'resnet': |
| 87 | if self.depth not in DepthNet.__factory: |
| 88 | raise KeyError("Unsupported depth:", self.depth) |
| 89 | self.encoder = DepthNet.__factory[depth](pretrained=self.pretrained) |
| 90 | elif self.backbone == 'resnext101_32x8d': |
| 91 | self.encoder = Resnext_torch.resnext101_32x8d(pretrained=self.pretrained) |
| 92 | else: |
| 93 | self.encoder = Resnext_torch.resnext101(pretrained=self.pretrained) |
| 94 | |
| 95 | def forward(self, x): |
| 96 | x = self.encoder(x) # 1/32, 1/16, 1/8, 1/4 |
| 97 | return x |
| 98 | |
| 99 | |
| 100 | class FTB(nn.Module): |
no outgoing calls
no test coverage detected