mmagic.models.editors.arcface.arcface_modules

Module Contents

Classes

Flatten	Flatten Module.
Bottleneck	A named tuple describing a ResNet block.
SEModule	Squeeze-and-Excitation Modules.
bottleneck_IR	Intermediate Resblock of bottleneck.
bottleneck_IR_SE	Intermediate Resblock of bottleneck with SEModule.

Functions

l2_norm(input[, axis])	l2 normalization.
get_block(in_channel, depth, num_units[, stride])	Get a single block config.
get_blocks(num_layers)	Get block configs of backbone.

class mmagic.models.editors.arcface.arcface_modules.Flatten

Bases: torch.nn.Module

Flatten Module.

forward(input)

mmagic.models.editors.arcface.arcface_modules.l2_norm(input, axis=1)

l2 normalization.

Parameters

• input (torch.Tensor) – The input tensor.

• axis (int, optional) – Specifies which axis of input to calculate the norm across. Defaults to 1.

Returns

Tensor after L2 normalization per-instance.

Return type

Tensor

class mmagic.models.editors.arcface.arcface_modules.Bottleneck

Bases: namedtuple('Block', ['in_channel', 'depth', 'stride'])

A named tuple describing a ResNet block.

mmagic.models.editors.arcface.arcface_modules.get_block(in_channel, depth, num_units, stride=2)

Get a single block config.

Parameters

• in_channel (int) – Input channels.

• depth (int) – Output channels.

• num_units (int) – Number of unit modules.

• stride (int, optional) – Conv2d stride. Defaults to 2.

Returns

A list of unit modules’ config.

Return type

list

mmagic.models.editors.arcface.arcface_modules.get_blocks(num_layers)

Get block configs of backbone.

Parameters

num_layers (int) – Number of ConvBlock layers in backbone.

Raises

ValueError – num_layers must be one of [50, 100, 152].

Returns

A list of block configs.

Return type

list

class mmagic.models.editors.arcface.arcface_modules.SEModule(channels, reduction)

Bases: torch.nn.Module

Squeeze-and-Excitation Modules.

Parameters

• channels (int) – Input channels.

• reduction (int) – Intermediate channels reduction ratio.

forward(x)

Forward Function.

class mmagic.models.editors.arcface.arcface_modules.bottleneck_IR(in_channel, depth, stride)

Bases: torch.nn.Module

Intermediate Resblock of bottleneck.

Parameters

• in_channel (int) – Input channels.

• depth (int) – Output channels.

• stride (int) – Conv2d stride.

forward(x)

Forward function.

class mmagic.models.editors.arcface.arcface_modules.bottleneck_IR_SE(in_channel, depth, stride)

Bases: torch.nn.Module

Intermediate Resblock of bottleneck with SEModule.

Parameters

• in_channel (int) – Input channels.

• depth (int) – Output channels.

• stride (int) – Conv2d stride.

forward(x)

Forward function.