r"""A Wrapper class that wraps ``torch.nn.Conv2d`` with normalization and nonlinearity. Args: in_channels (int): Number of channels i

r"""A Wrapper class that wraps ``torch.nn.Linear`` with normalization and nonlinearity. Args: in_features (int): Number of channels i

Dict as attribute trick.

r"""Residual block for 2D input. Args: in_channels (int) : Number of channels in the input tensor. out_channels (int) : Number of

r"""Configuration class. This should include every human specifiable hyperparameter values for your training.

r"""GAN loss constructor. Args: gan_mode (str): Type of GAN loss. ``'hinge'``, ``'least_square'``, ``'non_saturated'``, ``'wa

r"""The network that extracts features to compute the perceptual loss. Args: network (nn.Sequential) : The network that extracts features

r"""Perceptual loss initialization. Args: network (str) : The name of the loss network: 'vgg16' | 'vgg19'. layers (str or list of s

r"""Embed the input label map to get embedded features. Args: emb_cfg (obj): Embed network configuration. num_input_channels (int

Subclass torch's LayerNorm to handle fp16.

r"""Linear layer with affine modulation (Based on StyleGAN2 mod demod). Equivalent to affine modulation following linear, but faster when the same

InceptionC block patched for FID computation

r"""Compute feature matching loss

r"""Improved FUNIT style decoder. Args: input_dim (int): Input dimension (style code dimension). output_dim (int): Output dimensi

r"""Patch Discriminator constructor. Args: kernel_size (int): Convolution kernel size. num_input_channels (int): Num. of channels

r"""Global residual discriminator. Args: image_channels (int): Num. of channels in the real/fake image. num_filters (int): Num. o

r"""Add Gaussian noise to the input tensor.

Class to colorize segmentation maps.

InceptionA block patched for FID computation

r"""Compute KL loss in VAE for Gaussian distributions

Meter is to keep track of statistics along steps. Meters write values for purpose like printing average values. Meters can be flushed to log f

r"""Multi-resolution patch discriminator. Args: dis_cfg (obj): Discriminator part of the yaml config file. num_input_channels (in

r"""Multi-resolution patch discriminator. Args: num_discriminators (int): Num. of discriminators (one per scale). kernel_size (in

r"""Improved UNIT autoencoder. Args: num_filters (int): Base filter numbers. max_num_filters (int): Maximum number of filters in

r"""Improved MUNIT autoencoder. Args: num_filters (int): Base filter numbers. max_num_filters (int): Maximum number of filters in

r"""Improved FUNIT Content Encoder. This is basically the same as the original FUNIT content encoder. Args: num_downsamples (int): Nu

r"""Improved UNIT encoder. The network consists of: - input layers - $(num_downsamples) convolutional blocks - $(num_res_blocks) residual

r"""Improved FUNIT decoder. Args: num_enc_output_channels (int): Number of content feature channels. style_channels (int): Dimens

r"""flow generator constructor. Args: flow_cfg (obj): Flow definition part of the yaml config file. data_cfg (obj): Data definition

r"""Masked L1 loss constructor.

r"""In this model average implementation, the spectral layers are absorbed in the model parameter by default. If such options are turned on, b

r"""SPADE Image Generator constructor. Args: num_labels (int): Number of different labels. out_image_small_side_size (int): min(w

r"""Improved FUNIT Style Encoder. This is basically the same as the original FUNIT Style Encoder. Args: num_downsamples (int): Number

r"""Handles all weight reshape related tasks.

r"""Multi-resolution patch discriminator with shared weights. Args: num_discriminators (int): Num. of discriminators (one per scale).

r"""Dummy wrapping the module.

r"""Adaptive normalization layer. The layer first normalizes the input, then performs an affine transformation using parameters computed from the

r"""Learning affine modulation of activation. Args: in_features (int): Number of input features. style_features (int): Number of

r"""Attention module constructor. Args: atn_cfg (obj): Generator definition part of the yaml config file. data_cfg (obj): Data defi

r"""Handles data augmentation using albumentations library.

r"""COCO-FUNIT Generator architecture. Args: num_filters (int): Base filter numbers. num_filters_mlp (int): Base filter number in

r"""This creates an on disk cache, which saves files as bytes. Args: root (str): Path to the cache dir. size_MB (float): Size of c

r"""Improved UNIT decoder. The network consists of: - $(num_res_blocks) residual blocks. - $(num_upsamples) residual blocks or convolutional

r"""Improved MUNIT decoder. The network consists of - $(num_res_blocks) residual blocks. - $(num_upsamples) residual blocks or convolutional

r"""Encoder for getting region-wise features for style control. Args: enc_cfg (obj): Encoder definition part of the yaml config file.

First InceptionE block patched for FID computation

Second InceptionE block patched for FID computation

r"""# Feature-Pyramid Semantics Embedding Discriminator. This is a copy of the discriminator in https://arxiv.org/pdf/1910.06809.pdf

r""" Args: num_filters (int): Base filter numbers. num_filters_mlp (int): Base filter number in the MLP module. style_

r"""Flow generator constructor. Args: flow_cfg (obj): Flow definition part of the yaml config file. data_cfg (obj): Data definition

r"""Flow loss constructor. Args: cfg (obj): Configuration.

r"""This deals with opening, and reading from an Folder dataset. Args: root (str): Path to the folder. metadata (dict): Containin

r"""Fromage optimizer implementation (https://arxiv.org/abs/2002.03432)

r"""Manages the SPADE/GauGAN model used to generate pseudo-GTs for training GANcraft. Args: gaugan_cfg (Config): SPADE configuration.

r"""Coarse generator constructor. This is the main generator in the pix2pixHD architecture. Args: gen_cfg (obj): Generator definition

r"""Hyper Conv2d initialization. Args: in_channels (int): Dummy parameter. out_channels (int): Dummy parameter. kernel_si

r"""Hyper residual block for 2D input. Args: in_channels (int) : Number of channels in the input tensor. out_channels (int) : Num

r"""Spatially Adaptive Normalization (SPADE) initialization. Args: num_features (int) : Number of channels in the input tensor. c

r"""This deals with opening, and reading from an LMDB dataset. Args: root (str): Path to the LMDB file.

r"""Layer Normalization as introduced in https://arxiv.org/abs/1607.06450. This is the usual way to apply layer normalization in CNNs. Not

r"""Local enhancer constructor. These are sub-networks that are useful when aiming to produce high-resolution outputs. Args: gen_cfg

r"""The multi-layer perceptron (MLP) that maps Gaussian style code to a feature vector that is given as the conditional input to AdaIN. Args:

r"""MADAM optimizer implementation (https://arxiv.org/abs/2006.14560)

A ResNet class that is similar to torchvision's but contains the following changes: - There are now 3 "stem" convolutions as opposed to 1, wi

A single linear layer used as placeholder for LPIPS learnt weights

r"""This deals with opening, and reading from an AWS S3 bucket. Args: root (str): Path to the AWS S3 bucket. aws_credentials_file

r"""CNN converting intermediate feature map to final image.