mmagic.models.editors.ttsr.ttsr_net

Module Contents

Classes

TTSRNet	TTSR network structure (main-net) for reference-based super-resolution.
SFE	Structural Feature Encoder.
CSFI2	Cross-Scale Feature Integration between 1x and 2x features.
CSFI3	Cross-Scale Feature Integration between 1x, 2x, and 4x features.
MergeFeatures	Merge Features. Merge 1x, 2x, and 4x features.

Attributes

_conv3x3_layer
_conv1x1_layer

mmagic.models.editors.ttsr.ttsr_net._conv3x3_layer

mmagic.models.editors.ttsr.ttsr_net._conv1x1_layer

class mmagic.models.editors.ttsr.ttsr_net.TTSRNet(in_channels, out_channels, mid_channels=64, texture_channels=64, num_blocks=(16, 16, 8, 4), res_scale=1.0, init_cfg=None)

Bases: mmengine.model.BaseModule

TTSR network structure (main-net) for reference-based super-resolution.

Paper: Learning Texture Transformer Network for Image Super-Resolution

Adapted from ‘https://github.com/researchmm/TTSR.git’ ‘https://github.com/researchmm/TTSR’ Copyright permission at ‘https://github.com/researchmm/TTSR/issues/38’.

Parameters

• in_channels (int) – Number of channels in the input image

• out_channels (int) – Number of channels in the output image

• mid_channels (int) – Channel number of intermediate features. Default: 64

• texture_channels (int) – Number of texture channels. Default: 64.

• num_blocks (tuple[int]) – Block numbers in the trunk network. Default: (16, 16, 8, 4)

• res_scale (float) – Used to scale the residual in residual block. Default: 1.

• init_cfg (dict, optional) – Initialization config dict.

forward(x, soft_attention, textures)

Forward function.

Parameters

• x (Tensor) – Input tensor with shape (n, c, h, w).

• soft_attention (Tensor) – Soft-Attention tensor with shape (n, 1, h, w).

• textures (Tuple[Tensor]) – Transferred HR texture tensors. [(N, C, H, W), (N, C/2, 2H, 2W), …]

Returns

Forward results.

Return type

Tensor

class mmagic.models.editors.ttsr.ttsr_net.SFE(in_channels, mid_channels, num_blocks, res_scale)

Bases: torch.nn.Module

Structural Feature Encoder.

Backbone of Texture Transformer Network for Image Super-Resolution.

Parameters

• in_channels (int) – Number of channels in the input image

• mid_channels (int) – Channel number of intermediate features

• num_blocks (int) – Block number in the trunk network

• res_scale (float) – Used to scale the residual in residual block. Default: 1.

forward(x)

Forward function.

Parameters

x (Tensor) – Input tensor with shape (n, c, h, w).

Returns

Forward results.

Return type

Tensor

class mmagic.models.editors.ttsr.ttsr_net.CSFI2(mid_channels)

Bases: torch.nn.Module

Cross-Scale Feature Integration between 1x and 2x features.

Cross-Scale Feature Integration in Texture Transformer Network for

Image Super-Resolution.

It is cross-scale feature integration between 1x and 2x features.

For example, conv2to1 means conv layer from 2x feature to 1x feature. Down-sampling is achieved by conv layer with stride=2, and up-sampling is achieved by bicubic interpolate and conv layer.

Parameters

mid_channels (int) – Channel number of intermediate features

forward(x1, x2)

Forward function.

Parameters

• x1 (Tensor) – Input tensor with shape (n, c, h, w).

• x2 (Tensor) – Input tensor with shape (n, c, 2h, 2w).

Returns

Output tensor with shape (n, c, h, w). x2 (Tensor): Output tensor with shape (n, c, 2h, 2w).

Return type

x1 (Tensor)

class mmagic.models.editors.ttsr.ttsr_net.CSFI3(mid_channels)

Bases: torch.nn.Module

Cross-Scale Feature Integration between 1x, 2x, and 4x features.

Cross-Scale Feature Integration in Texture Transformer Network for

Image Super-Resolution.

It is cross-scale feature integration between 1x and 2x features.

For example, conv2to1 means conv layer from 2x feature to 1x feature. Down-sampling is achieved by conv layer with stride=2, and up-sampling is achieved by bicubic interpolate and conv layer.

Parameters

mid_channels (int) – Channel number of intermediate features

forward(x1, x2, x4)

Forward function.

Parameters

• x1 (Tensor) – Input tensor with shape (n, c, h, w).

• x2 (Tensor) – Input tensor with shape (n, c, 2h, 2w).

• x4 (Tensor) – Input tensor with shape (n, c, 4h, 4w).

Returns

Output tensor with shape (n, c, h, w). x2 (Tensor): Output tensor with shape (n, c, 2h, 2w). x4 (Tensor): Output tensor with shape (n, c, 4h, 4w).

Return type

x1 (Tensor)

class mmagic.models.editors.ttsr.ttsr_net.MergeFeatures(mid_channels, out_channels)

Bases: torch.nn.Module

Merge Features. Merge 1x, 2x, and 4x features.

Final module of Texture Transformer Network for Image Super-Resolution.

Parameters

• mid_channels (int) – Channel number of intermediate features

• out_channels (int) – Number of channels in the output image

forward(x1, x2, x4)

Forward function.

Parameters

• x1 (Tensor) – Input tensor with shape (n, c, h, w).

• x2 (Tensor) – Input tensor with shape (n, c, 2h, 2w).

• x4 (Tensor) – Input tensor with shape (n, c, 4h, 4w).

Returns

Output tensor with shape (n, c_out, 4h, 4w).

Return type

x (Tensor)