Logs a string to either file or console Args: s : str string to log filepath output filepath for logging

Logs a string to either file or console Arg(s): s : str string to log filepath output filepath for l

Logs a string to either file or console Arg(s): s : str string to log filepath output filepath for l

Moves model to specified device Arg(s): device : torch.device device for running model

Moves model to specified device Arg(s): device : torch.device device for running model

Move model to a device Arg(s): device : torch.device device to use

Returns the list of parameters in the model Returns: list : list of parameters

Applies transform to images and ground truth Arg(s): images_arr : list[torch.Tensor] list of N x C x H x

Performs cropping on images Arg(s): images_arr : list[torch.Tensor] list of N x C x H x W tensors

Sets model to evaluation mode

Forward input x through a convolution layer Arg(s): x : torch.Tensor[float32] N x C x H x W input tensor

Loads a network checkpoint file. Can be called in three different ways: load_checkpoint(): Loads the latest epoch

Returns the list of parameters in the model Returns: list : list of parameters

Removes erroneous measurements from sparse depth and validity map Arg(s): sparse_depth : torch.Tensor[float32]

Applies transform to images and ground truth Arg(s): images_arr : list[torch.Tensor] list of N x C x H x

input_tensor: (batch, x_dim, y_dim, c)

Sets model to evaluation mode

Removes erroneous measurements from sparse depth and validity map Arg(s): sparse_depth : torch.Tensor[float32]

Restore weights of the model Arg(s): checkpoint_path : str path to checkpoint optimizer : to

3x3 convolution with padding

Sets model to evaluation mode

Forwards the inputs through the network Arg(s): image0 : torch.Tensor[float32] image at time step 0

Forwards the inputs through the network Arg(s): image : torch.Tensor[float32] N x 3 x H x W image

Forwards the inputs through the network Arg(s): sparse_depth : torch.Tensor[float32] N x 1 x H x W spars

Load in triplet frames from path Arg(s): path : str path to image triplet normalize : bool if set, n

Logs summary to Tensorboard Arg(s): summary_writer : SummaryWriter Tensorboard summary writer

Returns the list of parameters in the model Returns: list : list of parameters

Loads weights from checkpoint Arg(s): restore_path : str path to model weights optimizer : t

Restore weights of the model Arg(s): checkpoint_path : str path to checkpoint optimizer : to

Restore weights of the model Arg(s): checkpoint_path : str path to checkpoint optimizer : to

Save weights of the model to checkpoint path Arg(s): checkpoint_path : str path to save checkpoint

Save weights of the model to checkpoint path Arg(s): checkpoint_path : str path to save checkpoint

Do training for the given number of epochs.

Sets model to evaluation mode

Perform horizontal flip on each sample Arg(s): images_arr : list[torch.Tensor[float32]] list of N x C x

Perform horizontal flip on each sample Arg(s): images_arr : list[torch.Tensor[float32]] list of N x C x

Perform horizontal flip on each sample Arg(s): images_arr : list[torch.Tensor[float32]] list of N x C x

Load in triplet frames from path Arg(s): path : str path to image triplet normalize : bool if set, n

Returns the set of parameters

Apply crop to inputs e.g. images, depth and if available adjust camera intrinsics Arg(s): inputs : list[numpy[float32]]

Removes erroneous measurements from sparse depth and validity map Arg(s): sparse_depth : torch.Tensor[float32]

Restore weights of the model Arg(s): checkpoint_path : str path to checkpoint optimizer : to

Runs an external depth completion model if output paths are provided, then will save outputs to a predetermined list of paths Arg(s):

Saves a checkpoint of the network and other variables.

Applies transform to images and ground truth Arg(s): images_arr : list[torch.Tensor] list of N x C x H x

Perform vertical flip on each sample Arg(s): images_arr : list[torch.Tensor[float32]] list of N x C x H

Perform vertical flip on each sample Arg(s): images_arr : list[torch.Tensor[float32]] list of N x C x H

Perform vertical flip on each sample Arg(s): images_arr : list[torch.Tensor[float32]] list of N x C x H

Computes loss function l = w_{ph}l_{ph} + w_{sz}l_{sz} + w_{sm}l_{sm} + w_{tp}l_{tp} Arg(s): output_depth0 : tor

1x1 convolution

Allows multi-gpu split along batch

Allows multi-gpu split along batch