Gather-and-Expansion Layer. Args: in_channels (int): Number of input channels. out_channels (int): Number of output channels.

Image to Patch Embedding. We use a conv layer to implement PatchEmbed. Args: in_channels (int): The num of input channels. Default:

InvertedResidual block for MobileNetV2. Args: in_channels (int): The input channels of the InvertedResidual block. out_channels (

r"""LayerNorm that supports two data formats: channels_last (default) or channels_first. The ordering of the dimensions in the inputs. channels_la

Linear Embedding:

Linear Embedding:

Pooling Pyramid Module used in PSPNet. Args: pool_scales (tuple[int]): Pooling scales used in Pooling Pyramid Module.

Attention Refinement Module (ARM) to refine the features of each stage. Args: in_channels (int): The number of input channels. ou

r"""LayerNorm that supports two data formats: channels_last (default) or channels_first. The ordering of the dimensions in the inputs. channels_la

Linear Embedding:

Non-bottleneck block of ERFNet. Args: channels (int): Number of channels in Non-bottleneck block. drop_rate (float): Probability

ResLayer to build ResNet style backbone. Args: block (nn.Module): block used to build ResLayer. inplanes (int): inplanes of block

Applies padding to input (if needed) so that input can get fully covered by filter you specified. It support two modes "same" and "corner". The

Cascade Feature Fusion Unit in ICNet. Args: low_channels (int): The number of input channels for low resolution feature map.

Context Guided Block for CGNet. This class consists of four components: local feature extractor, surrounding feature extractor, joint feature

Downsampler block of ERFNet. This module is a little different from basical ConvModule. The features from Conv and MaxPool layers are con

Downsampling module for CGNet.

2D Non-local module. Args: in_channels (int): Same as `NonLocalND`. sub_sample (bool): Whether to apply max pooling after pairwis

ResidualConvUnit, pre-activate residual unit. Args: in_channels (int): number of channels in the input feature map. act_cfg (dict

Self-Attention Module. Args: in_channels (int): Input channels of key/query feature. channels (int): Output channels of key/query

Make a ANN used SelfAttentionBlock. Args: low_in_channels (int): Input channels of lower level feature, which is the key feat

Implements one encoder layer in Vision Transformer. Args: embed_dims (int): The feature dimension. num_heads (int): Parallel atte

Upsampler block of ERFNet. Args: in_channels (int): Number of input channels. out_channels (int): Number of output channels.

Adaptive Context Module used in APCNet. Args: pool_scale (int): Pooling scale used in Adaptive Context Module to extract regi

Asymmetric Fusion Non-local Block(AFNB) Args: low_in_channels (int): Input channels of lower level feature, which is the key

Asymmetric Pyramid Non-local Block (APNB) Args: in_channels (int): Input channels of key/query feature, which is the key feat

Atrous Spatial Pyramid Pooling (ASPP) Module. Args: dilations (tuple[int]): Dilation rate of each layer. in_channels (int): Input

Window based multi-head self-attention (W-MSA) module with relative position bias. Args: embed_dims (int): Number of input channels.

Implements one encoder layer in Vision Transformer. Args: embed_dims (int): The feature dimension. num_heads (int): Parallel atte

Bilateral Guided Aggregation Layer to fuse the complementary information from both Detail Branch and Semantic Branch. Args: out_chann

Basic convolutional block for UNet. This module consists of several plain convolutional layers. Args: in_channels (int): Number of i

A basic RGB-D layer in DFormerv2.

Channel Attention Module (CAM)

Context Embedding Block for large receptive filed in Semantic Branch. Args: in_channels (int): Number of input channels. Defa

The Conditional Position Encoding (CPE) module. The CPE is the implementation of 'Conditional Positional Encodings for Vision Transformers <h

Context Path to provide sufficient receptive field. Args: backbone_cfg:(dict): Config of backbone of Context Path. co

Dynamic Convolutional Module used in DMNet. Args: filter_size (int): The filter size of generated convolution kernel used in

Atrous Spatial Pyramid Pooling (ASPP) Module with depthwise separable conv.

Detail Branch with wide channels and shallow layers to capture low-level details and generate high-resolution feature representation. Args:

Disentangled Non-Local Blocks. Args: temperature (float): Temperature to adjust attention. Default: 0.05

Expectation Maximization Attention Module used in EMANet. Args: channels (int): Channels of the whole module. num_bases (int): Nu

An implementation of Efficient Multi-head Attention of Segformer. This module is modified from MultiheadAttention which is a module from mmcv

Encoding Module used in EncNet. Args: in_channels (int): Input channels. num_codes (int): Number of code words. conv_cfg

Encoding Layer: a learnable residual encoder. Input is of shape (batch_size, channels, height, width). Output is of shape (batch_size, num_c

FeatureFusionBlock, merge feature map from different stages. Args: in_channels (int): Input channels. act_cfg (dict): The activat

Feature Fusion Module. This module is different from FeatureFusionModule in BiSeNetV1. It uses two ConvModules in `self.attention` whose inter

Feature Fusion Module to fuse low level output feature of Spatial Path and high level output feature of Context Path. Args: in_channe

Feature fusion module. Args: higher_in_channels (int): Number of input channels of the higher-resolution branch. lowe

Implements one encoder layer with GSA. Args: embed_dims (int): The feature dimension. num_heads (int): Parallel attention heads.

Global Context Extractor for CGNet. This class is employed to refine the joint feature of both local feature and surrounding context. Ar

Global feature extractor module. Args: in_channels (int): Number of input channels of the GFE module. Default: 64 blo

Global Sub-sampled Attention (Spatial Reduction Attention) This module is modified from EfficientMultiheadAttention， which is a module from m

High-Resolution Module for HRNet. In this module, every branch has 4 BasicBlocks/Bottlenecks. Fusion/Exchange is in this module.

Implements one encoder layer in Twins-SVT. Args: embed_dims (int): The feature dimension. num_heads (int): Parallel attention hea

Learning to downsample module. Args: in_channels (int): Number of input channels. dw_channels (tuple[int]): Number of output chan

A simple pipeline to load image.

Locally-grouped Self Attention (LSA) module. Args: embed_dims (int): Number of input channels. num_heads (int): Number of attenti

Multi-head self-attention with relative position bias used in MAE. This module is different from ``BEiTAttention`` by initializing the relati

Implements one encoder layer in Vision Transformer. This module is different from ``BEiTTransformerEncoderLayer`` by replacing ``BEiTAttentio

An implementation of MixFFN of Segformer. The differences between MixFFN & FFN: 1. Use 1X1 Conv to replace Linear layer. 2. Intro

Make a OCR used SelfAttentionBlock.

Position Attention Module (PAM) Args: in_channels (int): Input channels of key/query feature. channels (int): Output channels of

Pooling Pyramid Module used in PSPNet. Args: pool_scales (tuple[int]): Pooling scales used in Pooling Pyramid Module.

Pyramid Pooling Module that only concat the features of each layer. Args: pool_scales (tuple[int]): Pooling scales used in Pooling Pyrami

Image to Patch Embedding

Merge patch feature map. This layer groups feature map by kernel_size, and applies norm and linear layers to the grouped feature map. Our imp

Radix Softmax module in ``SplitAttentionConv2d``. Args: radix (int): Radix of input. groups (int): Groups of input.

ViTPostProcessBlock, process cls_token in ViT backbone output and rearrange the feature vector to feature map. Args: in_channels (int

Squeeze-and-Excitation Module. Args: channels (int): The input (and output) channels of the SE layer. ratio (int): Squeeze ratio