RMSNorm layer.

A conv block that bundles conv/norm/activation layers. This block simplifies the usage of convolution layers, which are commonly used with a

Computes multi-head attention. Supports nested or padded tensors. Args: E_q (int): Size of embedding dim for query E_k (int)

Augmentation pipeline that supports deterministic replay. Parameters ---------- All parameters are identical to the original NVIDIA imple

Encoder Decoder depther. EncoderDecoder typically consists of backbone and decode_head.

Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).

Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).

Upsampler.

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

Residual convolution module.

Just a batchnorm.

Vision Transformers for Dense Prediction. This head is implemented of `DPT <https://arxiv.org/abs/2103.13413>`_. Args: embed_dims (int

Dino v2 wrapper using original implementation, hacked with modulation.

Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).

Feature fusion block.

FeatureFusionBlock, merge feature map from different stages. Args: in_channels (int): Input channels. act_layer (nn.Module): activ

Fast gelu implementation.

Normlize image by given mean and std.

Upsampling block.

Prepare sample for usage as network input.

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

Resize sample to given size (width, height).

Base class for BaseDecodeHead. Args: in_channels (List): Input channels. channels (int): Channels after modules, before conv_dept

Normlize image by given mean and std.

2D image to patch embedding: (B,C,H,W) -> (B,N,D) Args: img_size: Image size. patch_size: Patch token size. in_chans

2D image to patch embedding: (B,C,H,W) -> (B,N,D) Args: img_size: Image size. patch_size: Patch token size. in_chans

Prepare sample for usage as network input.

Attention block.

Resize sample to given size (width, height).