Combined Swin Transformer and Convolution Block.

NAFNet building block with simplified channel attention and SimpleGate.

Channel-wise layer normalization for 2D feature maps.

Swin Transformer Block.

Stochastic depth (drop path) for regularization.

NAFNet: Nonlinear Activation Free Network for image restoration. Encoder-decoder architecture with skip connections, using NAFBlock as the ba

SCUNet: Swin-Conv-UNet for blind image denoising. Args: in_nc: Number of input channels. Default: 3. config: Number of ConvTransB

Split channels in half and multiply - a simple gating mechanism.

Window Multi-head Self-Attention module in Swin Transformer.