mmagic.models.editors.cain.cain_net

Module Contents

Classes

CAINNet	CAIN network structure.
ConvNormWithReflectionPad	Apply reflection padding, followed by a convolution, which can be
ChannelAttentionLayer	Channel Attention (CA) Layer.
ResidualChannelAttention	Residual Channel Attention Module.
ResidualGroup	Residual Group, consisting of a stack of residual channel attention,

Functions

get_padding_functions(x[, padding])

Generate padding function for CAIN.

class mmagic.models.editors.cain.cain_net.CAINNet(in_channels=3, kernel_size=3, num_block_groups=5, num_block_layers=12, depth=3, reduction=16, norm=None, padding=7, act=nn.LeakyReLU(0.2, True), init_cfg=None)

Bases: mmengine.model.BaseModule

CAIN network structure.

Paper: Channel Attention Is All You Need for Video Frame Interpolation. Ref repo: https://github.com/myungsub/CAIN

Parameters

• in_channels (int) – Channel number of inputs. Default: 3.

• kernel_size (int) – Kernel size of CAINNet. Default: 3.

• num_block_groups (int) – Number of block groups. Default: 5.

• num_block_layers (int) – Number of blocks in a group. Default: 12.

• depth (int) – Down scale depth, scale = 2**depth. Default: 3.

• reduction (int) – Channel reduction of CA. Default: 16.

• norm (str | None) – Normalization layer. If it is None, no normalization is performed. Default: None.

• padding (int) – Padding of CAINNet. Default: 7.

• act (function) – activate function. Default: nn.LeakyReLU(0.2, True).

• init_cfg (dict, optional) – Initialization config dict. Default: None.

forward(imgs, padding_flag=False)

Forward function.

Parameters

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

• padding_flag (bool) – Padding or not. Default: False.

Returns

Forward results.

Return type

Tensor

mmagic.models.editors.cain.cain_net.get_padding_functions(x, padding=7)

Generate padding function for CAIN.

This function produces two functions to pad and depad a tensor, given the number of pixels to be padded. When applying padding and depadding sequentially, the original tensor is obtained.

The generated padding function will pad the given tensor to the ‘padding’ power of 2, i.e., pow(2, ‘padding’).

tensor –padding_function–> padded tensor padded tensor –depadding_function–> original tensor

Parameters

• x (Tensor) – Input tensor.

• padding (int) – Padding size. Default: 7.

Returns

Padding function. depadding_function (Function): Depadding function.

Return type

padding_function (Function)

class mmagic.models.editors.cain.cain_net.ConvNormWithReflectionPad(in_channels, out_channels, kernel_size, norm=None)

Bases: mmengine.model.BaseModule

Apply reflection padding, followed by a convolution, which can be followed by an optional normalization.

Parameters

• in_channels (int) – Channel number of input features.

• out_channels (int) – Channel number of output features.

• kernel_size (int) – Kernel size of convolution layer.

• norm (str | None) – Normalization layer. If it is None, no normalization is performed. Default: None.

forward(x)

Forward function for ConvNormWithReflectionPad.

Parameters

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

Returns

Output tensor with shape (n, c, h, w).

Return type

Tensor

class mmagic.models.editors.cain.cain_net.ChannelAttentionLayer(mid_channels, reduction=16)

Bases: mmengine.model.BaseModule

Channel Attention (CA) Layer.

Parameters

• mid_channels (int) – Channel number of the intermediate features.

• reduction (int) – Channel reduction of CA. Default: 16.

forward(x)

Forward function for ChannelAttentionLayer.

Parameters

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

Returns

Output tensor with shape (n, c, h, w).

Return type

Tensor

class mmagic.models.editors.cain.cain_net.ResidualChannelAttention(mid_channels, kernel_size=3, reduction=16, norm=None, act=nn.LeakyReLU(0.2, True))

Bases: mmengine.model.BaseModule

Residual Channel Attention Module.

Parameters

• mid_channels (int) – Channel number of the intermediate features.

• kernel_size (int) – Kernel size of convolution layers. Default: 3.

• reduction (int) – Channel reduction. Default: 16.

• norm (None | function) – Norm layer. If None, no norm layer. Default: None.

• act (function) – activation function. Default: nn.LeakyReLU(0.2, True).

forward(x)

Forward function for ResidualChannelAttention.

Parameters

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

Returns

Output tensor with shape (n, c, h, w).

Return type

Tensor

class mmagic.models.editors.cain.cain_net.ResidualGroup(block_layer, num_block_layers, mid_channels, kernel_size, reduction, act=nn.LeakyReLU(0.2, True), norm=None)

Bases: mmengine.model.BaseModule

Residual Group, consisting of a stack of residual channel attention, followed by a convolution.

Parameters

• block_layer (nn.Module) – nn.Module class for basic block.

• num_block_layers (int) – number of blocks.

• mid_channels (int) – Channel number of the intermediate features.

• kernel_size (int) – Kernel size of ResidualGroup.

• reduction (int) – Channel reduction of CA. Default: 16.

• act (function) – activation function. Default: nn.LeakyReLU(0.2, True).

• norm (str | None) – Normalization layer. If it is None, no normalization is performed. Default: None.

forward(x)

Forward function for ResidualGroup.

Parameters

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

Returns

Output tensor with shape (n, c, h, w).

Return type

Tensor