r"""Adds an LMDB dataset to a list of datasets. Args: root (str): Path to LMDB or folder dataset. filenames: List of

r"""Assign training loss terms. Args: name (str): Loss name criterion (obj): Loss object. weight (float):

r"""Return data loader . Args: cfg (obj): Global configuration file. dataset (obj): PyTorch dataset object. batch_size (i

r"""Download object from bucket. Args: key (str): Key inside bucket.

r"""Convert numpy to torch tensor. Args: normalize (bool): Normalize image i.e. (x - 0.5) * 2. Goes from [0, 1] -

Lanczos algorithm computes symmetric m x m tridiagonal matrix T and matrix V with orthogonal rows constituting the basis of the Krylov su

r"""Compute the polynomial kernel between X and Y

r""" Record time before applying backward.

r""" Record time before updating the weights

r"""Normalize using ImageNet mean and std. Args: input (4D tensor NxCxHxW): The input images, assuming to be [-1, 1]. Returns:

r"""Main function. Args: data_root (str): Location of dataset root. cfg (object): Loaded config object. paired (bool): Pa

r"""Discriminate individual images. Args: net_D (obj): Discriminator network. real_A (NxC1xHxW tensor): Input label m

r"""Get the dimensions of conditional inputs. Args: num_downs (int) : How many downsamples at current layer. Returns:

r"""Get the conditional inputs. Args: label (4D tensor) : Input label tensor. embedder (obj) : Embedding network.

r"""Get the conditional inputs. Args: label (4D tensor) : Input label tensor. embedder (obj) : Embedding network.

r"""Get data at current time frame given the sequence of data. Args: data (dict): Training data for current iteration.

r"""Iteratively try to get the attribute from cfg. If not found, return default. Args: cfg (obj): Config file. attr_name (str

r"""Return the scheduler object. Args: cfg_opt (obj): Config for the specific optimization module (gen/dis). net (obj): PyTorch n

r"""Return the target vector for the binary cross entropy loss computation. Args: dis_output (tensor): Discriminator outp

r"""Load data item stored for key = path. Args: path (str): Key into LMDB dataset. data_type (str): Key into self.ext

r"""Initialize the cudnn module. The two things to consider is whether to use cudnn benchmark and whether to use cudnn deterministic. If cudnn

r"""Initialize distributed training

r"""Create log directory for storing checkpoints and output images. Args: config_path (str): Path to the configuration file. logd

r"""Number of sequences available for inference. Returns: (int)

r"""Pre-process the DensePose part of input label map. Args: pose_cfg (obj): Pose data configuration. pose_map (NxCxHxW tensor):

r"""Save network weights, optimizer parameters, scheduler parameters to a checkpoint.

r"""Set GPU affinity Args: gpu_id (int): Which gpu device.

r"""Convenience function to setup 2D FIR filter for `upfirdn2d()`. Args: f: Torch tensor, numpy array, or python list of the sh

r"""The inference function. If output_dir exists, also save the output image. Args: data (dict): Training data at the curr

r"""Calculate number of parameters in a PyTorch network. Args: model (obj): PyTorch network. Returns: (int): Number of param

r"""Compute FID values.

r"""We will compute FID for the regular model using the eval mode. For the moving average model, we will use the eval mode.

r"""Create the LMDB key for this piece of information. Args: sequence_name (str): Which sequence from the chosen dataset.

r"""Detach all logging variables to prevent potential memory leak.

r"""Forward the CNN Args: net_out (N x C5 x H x W tensor): Intermediate feature maps. z (N x C3 tensor): Intermediate

r"""Sample points along rays, forwarding the per-point MLP and aggregate pixel features Args: blk_feats (K x C1 tensor): Sparse b

r"""Load single image model, if any.

r"""Perform same data augmentation on all inputs. Args: inputs (dict): Keys are from self.augmentable_data_types. Values are

r"""Computes MMD between two sets of features using polynomial kernels. It performs a number of repetitions of subset sampling without replacement

r"""Load data and perform augmentation. Args: keys (dict): Key into LMDB/folder dataset for this item. Returns:

r"""Makes arrays bigger, if needed. Args: max_point_idx (int): Highest 3D point index seen so far.

r""" Record time before applying forward.

Fast CUDA implementation of `upfirdn2d()` using custom ops.

r"""Wrap a model for distributed data parallel training. Args: model (obj): PyTorch network model. Returns: (obj): Wrapped P

r"""Encode the input image to get the attention map. Args: img (NxCxHxW tensor): Input image. net_name (str): Name fo

r"""Convert OpenPose dicts to numpy arrays of keypoints. Args: inputs (list of dicts): List of OpenPose dicts. return_largest_onl

r"""Compute losses on the generated flow map. Args: flow (tensor): Generated flow map. warped_image (tensor): Warped

r"""Compute losses on the generated occlusion mask. Args: occ_mask (tensor): Generated occlusion mask. warped_image (

r"""Connect the face keypoints to edges and draw the sketch. Args: resize_h (int): Height the input image was resized to. resize_

r"""The update is already done in gen_update. Args: data (dict): Training data at the current iteration.

r""" Reduce to all ranks

r"""Download a file from google drive or pbss by using the url. Args: URL: GDrive URL or PBSS pre-signed URL for the checkpoint.

r"""Hyper Conv2d forward. Convolve x using the provided weight and bias. Args: x (N x C x H x W tensor): Input tensor.

r"""Quadratic fitting function.

r"""Generate a sequence of frames given a sequence of data. Args: data (dict): Training data at the current iteration.

r"""Update the vid2vid generator. We update in the fashion of dis_update (frame 1), gen_update (frame 1), dis_update (frame 2), gen_up

r"""Compute activation values and pack them in a list. Args: data_loader (obj): PyTorch dataloader object. key_real (str): Dictio

r"""Get data at current time frame given the sequence of data. Args: data (dict): Training data for current iteration.

r"""Get the bbox coordinates for face region. Args: keypoints (Nx2 tensor): Facial landmarks. orig_img_size (int tuple): Height a

r"""Obtain mask of faces. Args: densepose_map (3D or 4D tensor): DensePose map. Returns: mask (3D or 4D tensor): Face mask.

r"""Get mean distances of the points from face center.

r"""Get the mean xy distances of keypoints wrt face center.

r"""Get function to apply for specific op. Args: op (str): Name of the op. Returns: function handle.

r"""Return the scheduler object. Args: cfg_opt (obj): Config for the specific optimization module (gen/dis). params (obj): Parame

r"""Compute activation values and pack them in a list. Args: data_loader (obj): PyTorch dataloader object. key_real (str): Dictio

r"""Obtain mask of different body parts of humans. This is done by looking at the body part map from DensePose. Args: densepose_map (

r"""Get the bbox (pixel coordinates) to crop for person body region. Args: pose_map (NxCxHxW tensor): Input pose map. orig_img_si

Return the scheduler object. Args: cfg_opt (obj): Config for the specific optimization module (gen/dis). opt (obj): PyTorch optim

r"""Return dataset objects for the training and validation sets. Args: cfg (obj): Global configuration file. Returns: (dict)

r"""Return weight normalization. Args: norm_type (str): Type of weight normalization. ``'none'``, ``'spectral'``,

r"""UNIT inference. Args: data (dict): Training data at the current iteration. - images_a (tensor): Images from dom

r"""Initialize weights in net_dst with those in net_src.

r"""Given the start and end points, interpolate to get a curve/line. Args: x (1D array): x coordinates of the points to interpolate.

r"""N+1 label GAN loss computation. Args: input_x (tensor): Output values. t_real (boolean): Is this output value for