PackNet01

class packnet_sfm.networks.depth.PackNet01.PackNet01(dropout=None, version=None, **kwargs)[source]

Bases: torch.nn.modules.module.Module

PackNet network with 3d convolutions (version 01, from the CVPR paper).

https://arxiv.org/abs/1905.02693

Parameters
  • dropout (float) – Dropout value to use

  • version (str) – Has a XY format, where: X controls upsampling variations (not used at the moment). Y controls feature stacking (A for concatenation and B for addition)

  • kwargs (dict) – Extra parameters

forward(x)[source]

Runs the network and returns inverse depth maps (4 scales if training and 1 if not).

init_weights()[source]

Initializes network weights.