PoseResNet

class packnet_sfm.networks.pose.PoseResNet.PoseResNet(version=None, **kwargs)[source]

Bases: torch.nn.modules.module.Module

Pose network based on the ResNet architecture.

Parameters
  • version (str) – Has a XY format, where: X is the number of residual layers [18, 34, 50] and Y is an optional ImageNet pretrained flag added by the “pt” suffix Example: “18pt” initializes a pretrained ResNet18, and “34” initializes a ResNet34 from scratch

  • kwargs (dict) – Extra parameters

forward(target_image, ref_imgs)[source]

Runs the network and returns predicted poses (1 for each reference image).