direct.nn.didn package

Submodules

direct.nn.didn.didn module

class direct.nn.didn.didn.Subpixel(in_channels, out_channels, upscale_factor, kernel_size, padding=0)

Bases: Module

Subpixel convolution layer for up-scaling of low resolution features at super-resolution as implemented in [1]_.

References:

[1]

Yu, Songhyun, et al. “Deep Iterative Down-Up CNN for Image Denoising.” 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2019, pp. 2095–103. IEEE Xplore, https://doi.org/10.1109/CVPRW.2019.00262.

__init__(in_channels, out_channels, upscale_factor, kernel_size, padding=0)

Inits Subpixel.

Parameters:

• in_channels (int) – Number of input channels.

• out_channels (int) – Number of output channels.

• upscale_factor (int) – Subpixel upscale factor.

• kernel_size (Union[int, Tuple[int, int]]) – Convolution kernel size.

• padding (int) – Padding size. Default: 0.

forward(x)

Computes Subpixel convolution on input tensor.

Parameters:

x (Tensor) – Input tensor.

Return type:

Tensor

Returns:

Output tensor after subpixel convolution.

class direct.nn.didn.didn.ReconBlock(in_channels, num_convs)

Bases: Module

Reconstruction Block of DIDN model as implemented in [1]_.

References:

[1]

Yu, Songhyun, et al. “Deep Iterative Down-Up CNN for Image Denoising.” 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2019, pp. 2095–103. IEEE Xplore, https://doi.org/10.1109/CVPRW.2019.00262.

__init__(in_channels, num_convs)

Inits ReconBlock.

Parameters:

• in_channels (int) – Number of input channels.

• num_convs (int) – Number of convolution blocks.

forward(input_data)

Computes num_convs convolutions followed by PReLU activation on input tensor.

Parameters:

• input_data (Tensor) – torch.Tensor

• tensor. (Input)

Return type:

Tensor

class direct.nn.didn.didn.DUB(in_channels, out_channels)

Bases: Module

Down-up block (DUB) for DIDN model as implemented in [1]_.

References:

[1]

Yu, Songhyun, et al. “Deep Iterative Down-Up CNN for Image Denoising.” 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2019, pp. 2095–103. IEEE Xplore, https://doi.org/10.1109/CVPRW.2019.00262.

__init__(in_channels, out_channels)

Inits DUB.

Parameters:

• in_channels (int) – int

• channels. (Number of output)

• out_channels (int) – int

• channels.

static pad(x)

Pads input to height and width dimensions if odd.

Parameters:

• x (Tensor) – torch.Tensor

• pad. (Input to)

Returns:

torch.Tensor Padded tensor.

Return type:

x

static crop_to_shape(x, shape)

Crops x to specified shape.

Parameters:

• x (Tensor) – torch.Tensor

• shape (Tuple[int, int])

• shape – Tuple(int, int)

• H (Crop shape corresponding to)

• W.

Returns:

torch.Tensor Cropped tensor.

Return type:

cropped_output

forward(x)

Parameters:

• x (Tensor) – torch.Tensor

• tensor. (Input)

Returns:

torch.Tensor DUB output.

Return type:

out

class direct.nn.didn.didn.DIDN(in_channels, out_channels, hidden_channels=128, num_dubs=6, num_convs_recon=9, skip_connection=False)

Bases: Module

Deep Iterative Down-up convolutional Neural network (DIDN) implementation as in [1]_.

References:

[1]

Yu, Songhyun, et al. “Deep Iterative Down-Up CNN for Image Denoising.” 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), 2019, pp. 2095–103. IEEE Xplore, https://doi.org/10.1109/CVPRW.2019.00262.

__init__(in_channels, out_channels, hidden_channels=128, num_dubs=6, num_convs_recon=9, skip_connection=False)

Inits DIDN.

Parameters:

• in_channels (int) – int

• channels. (Number of output)

• out_channels (int) – int

• channels.

• hidden_channels (int) – int

• Default (Use skip connection.) –

• num_dubs (int) – int

• Default –

• num_convs_recon (int) – int

• Default –

• skip_connection (bool) – bool

• Default – False.

static crop_to_shape(x, shape)

Crops x to specified shape.

Parameters:

• x (Tensor) – torch.Tensor

• shape (Tuple[int, int])

• shape – Tuple(int, int)

• H (Crop shape corresponding to)

• W.

Returns:

torch.Tensor Cropped tensor.

Return type:

cropped_output

forward(x, channel_dim=1)

Takes as input a torch.Tensor x and computes DIDN(x).

Parameters:

• x (Tensor) – torch.Tensor

• tensor. (Input)

• channel_dim (int) – int

• Default (Channel dimension.)

Returns:

torch.Tensor DIDN output tensor.

Return type:

out

Module contents