Pixel Shuffle upsample layer. Args: in_channels (int): Number of input channels. out_channels (int): Number of output channels.

Compose a data pipeline with a sequence of transforms. Args: transforms (list[dict | callable]): Either config dicts of trans

Mask convolution module. This is a simple wrapper for mask convolution like: 'partial conv'. Convolutions in this module always need a mask a

Conv2d or Linear layer with max feature selector. Generate feature maps with double channels, split them and select the max feature.

Residual blocks with a convolution in front. Args: in_channels (int): Number of input channels of the first conv. out_channels (i

ResBlock for Hourglass. It has a style of: :: ---Conv-ReLU-Conv-Conv-+- |_________Conv________| or ---Co

Construct a Unet submodule with skip connections, with the following. structure: downsampling - `submodule` - upsampling. Args: oute

Equalized LR Linear Module with Activation Layer. Args: nn ([type]): [description]

Subclass torch's LayerNorm to handle fp16.

InceptionC block patched for FID computation.

Gatting of SE attention. Args: in_channels (int): Number of channels in the input feature map.

Apply reflection padding, followed by a convolution, which can be followed by an optional normalization. Args: in_channels (int): Cha

InceptionA block patched for FID computation.

Residual Dense Block. Used in RRDB block in ESRGAN. Args: mid_channels (int): Channel number of intermediate features. growt

A conv block that bundles conv/SEGating/norm layers. Args: in_channels (int): Number of channels in the input feature map. Sa

ASPP module from DeepLabV3. The code is adopted from https://github.com/pytorch/vision/blob/master/torchvision/models/ segmentation/deepl

A conv block that bundles conv/SEGating/norm layers. Args: in_channels (int): Number of channels in the input feature map. Sa

Depthwise separable convolution module. See https://arxiv.org/pdf/1704.04861.pdf for details. This module can replace a ConvModule with the

A DistributedDataParallel wrapper for models in MMediting. In MMedting, there is a need to wrap different modules in the models with separate

This class implements an image buffer that stores previously generated images. This buffer allows us to update the discriminator using a hist

Guided Contextual Attention Module. From https://arxiv.org/pdf/2001.04069.pdf. Based on https://github.com/nbei/Deep-Flow-Guided-Video-Inpain

Hourglass model for face landmark. It is a recursive model. Args: depth (int): Depth of Hourglass, the number of recursions.

Normalize images with the given mean and std value. Based on Conv2d layer, can work in GPU. Args: pixel_range (float): Pixel range o

Inverted residual layer for indexnet encoder. It basically is a depthwise separable conv module. If `expand_ratio` is not one, then a conv mo

r"""Modulated Conv2d in StyleGANv2. Attention: #. ``style_bias`` is provided to check the difference between official TF implementati

Modulated Deformable Convolutional Pack. Different from the official DCN, which generates offsets and masks from the preceding features, this

Multilayer Discriminator. This is a commonly used structure with stacked multiply convolution layers. Args: in_channels (int): Input

Normalize images with the given mean and std value. Required keys are the keys in attribute "keys", added or modified keys are the keys in at

Alignment module using Pyramid, Cascading and Deformable convolution (PCD). It is used in EDVRNet. Args: mid_channels (int): Number o

VGG network used in calculating perceptual loss. In this implementation, we allow users to choose whether use normalization in the input feat

Pretrained InceptionV3 network returning feature maps. This class is only used when TorchScript is not available, `torch.__version__` < '1.6.

Resize data to a specific size for training or resize the images to fit the network input regulation for testing. When used for resizing imag

SPyNet architecture. Note that this implementation is specifically for TOFlow. It differs from the common SPyNet in the following aspects:

SPyNet network structure. The difference to the SPyNet in [tof.py] is that 1. more SPyNetBasicModule is used in this version, and

Temporal Spatial Attention (TSA) fusion module. It is used in EDVRNet. Args: mid_channels (int): Number of the channels of middle feature

Convert some values in results dict to `torch.Tensor` type in data loader pipeline. Args: keys (Sequence[str]): Required keys to be c

AOT Block which constitutes the dilation backbone. This implementation follows: Aggregated Contextual Transformations for High-Resolution Ima

Augmented Deformable Convolution Pack. Different from DeformConv2dPack, which generates offsets from the preceding feature, this AugmentedDef

Basic module of SPyNet. Note that unlike the common spynet architecture, the basic module here contains batch normalization.

Basic module of SPyNet. Note that unlike the common spynet architecture, the basic module here could contain batch normalization. Args:

BasicVSR network structure for video super-resolution. Support only x4 upsampling. Paper: BasicVSR: The Search for Essential Componen

Cross-Scale Feature Integration between 1x and 2x features. Cross-Scale Feature Integration in Texture Transformer Network for Image Supe

Cross-Scale Feature Integration between 1x, 2x, and 4x features. Cross-Scale Feature Integration in Texture Transformer Network for Image

Channel Attention (CA) Layer. Args: mid_channels (int): Channel number of the intermediate features. reduction (int): Channel red

Contexture attention module. The details of this module can be found in: Generative Image Inpainting with Contextual Attention Args:

Decoder of FLAVR. Args: join_type (str): Join type of tensors from decoder and encoder. Candidates are ``concat`` and ``add``

Dense layer. Args: in_channels (int): Channel number of inputs. out_channels (int): Channel number of outputs.

Distributed evaluation hook. Args: dataloader (DataLoader): A PyTorch dataloader. interval (int): Evaluation interval. Default: 1

DistributedSampler inheriting from `torch.utils.data.DistributedSampler`. In pytorch of lower versions, there is no `shuffle` argument. This

EDVR feature extractor for information-refill in IconVSR. We use EDVR-M in IconVSR. To adopt pretrained models, please specify "pretrained".

Encoder of FLAVR. Args: block (nn.Module): Basic block of encoder. layers (str): List of layers in encoder. stem_layer (n

r"""Equalized Learning Rate. This trick is proposed in: Progressive Growing of GANs for Improved Quality, Stability, and Variation The g

r"""Equalized LR ConvModule. In this module, we inherit default ``mmcv.cnn.ConvModule`` and adopt equalized lr in convolution. The equalized

r"""Equalized LR LinearModule. In this module, we adopt equalized lr in ``nn.Linear``. The equalized learning rate is proposed in: Progre

Non-Distributed evaluation hook for iteration-based runner. This hook will regularly perform evaluation in a given interval when performing i

First InceptionE block patched for FID computation.

Second InceptionE block patched for FID computation.

Fusing Feature and Heatmap. Args: in_channels (int): Number of channels in the input features. num_heatmaps (int): Number of heat

Custom feedback block, will be used as the first feedback block. Args: in_channels (int): Number of channels in the input features.

Feedback block with HeatmapAttention. Args: in_channels (int): Number of channels in the input features. mid_channels (int): Numb

Feedback Hourglass model for face landmark. It has a style of: :: -- preprocessing ----- Hourglass ----->

A Gaussian filter which blurs a given tensor with a two-dimensional gaussian kernel by convolving it along each channel. Batch operation is su

Feature extractor features using InceptionV3 model. Args: style (str): The model style to run Inception model. it must be either

Indexed upsample module. Args: in_channels (int): Input channels. out_channels (int): Output channels. kernel_size (int,

LightCNN discriminator with input size 128 x 128. It is used to train DICGAN. Args: in_channels (int): Channel number of inputs.

Feature of LightCNN. It is used to train DICGAN.

A linear block that contains linear/norm/activation layers. For low level vision, we add spectral norm and padding layer. Args: in_f

Resize the input image using MATLAB-like downsampling. Currently support bicubic interpolation only. Note that the output of this function is

This module is modified from Pytorch `MaxUnpool2d` module. Args: kernel_size (int or tuple): Size of the max pooling window. stride (

Merge Features. Merge 1x, 2x, and 4x features. Final module of Texture Transformer Network for Image Super-Resolution. Args: mid_cha

Modified MiniBatch Stddev Layer. This layer is modified from ``MiniBatchStddevLayer`` used in PGGAN. In StyleGAN2, the authors add a new feat

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

r"""Modulated Conv2d in StyleGANv2. Attention: #. ``style_bias`` is provided to check the difference between official TF implementati

Pixel Normalization. This module is proposed in: Progressive Growing of GANs for Improved Quality, Stability, and Variation Args:

Residual Dense Block of Residual Dense Network. Args: in_channels (int): Channel number of inputs. channel_growth (int): Channels

Feature extractor composed of Residual-in-Residual Dense Blocks (RRDBs). It is equivalent to ESRGAN with the upsampling module removed. Args

A wrapper of repeated dataset. The length of repeated dataset will be `times` larger than the original dataset. This is useful when the data

ResNet architecture. Three-layers ResNet/ResBlock

Define a Residual Block with dropout layers. Ref: Deep Residual Learning for Image Recognition A residual block is a conv block with ski

Structural Feature Encoder. Backbone of Texture Transformer Network for Image Super-Resolution. Args: in_channels (int): Number of c

SPyNet architecture. Note that this implementation is specifically for TOFlow. It differs from the common SPyNet in the following aspects: