mmagic.models.editors.arcface.model_irse

Module Contents

Classes

Backbone

Arcface backbone.

Functions

IR_50(input_size)	Constructs a ir-50 model.
IR_101(input_size)	Constructs a ir-101 model.
IR_152(input_size)	Constructs a ir-152 model.
IR_SE_50(input_size)	Constructs a ir_se-50 model.
IR_SE_101(input_size)	Constructs a ir_se-101 model.
IR_SE_152(input_size)	Constructs a ir_se-152 model.

class mmagic.models.editors.arcface.model_irse.Backbone(input_size, num_layers, mode='ir', drop_ratio=0.4, affine=True)

Bases: torch.nn.Module

Arcface backbone. There are many repos follow this codes for facial recognition, and we also follow this routine. Ref: https://github.com/orpatashnik/StyleCLIP/blob/main/models/facial_recognition/helpers.py # noqa

Parameters

• input_size (int) – Input size of image.

• num_layers (int) – Number of layer in backbone.

• mode (str, optional) – Bottle neck mode. If set to ‘ir_se’, then SEModule will be applied. Defaults to ‘ir’.

• drop_ratio (float, optional) – Drop out ratio. Defaults to 0.4.

• affine (bool, optional) – Whether use affine in BatchNorm1d. Defaults to True.

forward(x)

Forward function.

mmagic.models.editors.arcface.model_irse.IR_50(input_size)

Constructs a ir-50 model.

mmagic.models.editors.arcface.model_irse.IR_101(input_size)

Constructs a ir-101 model.

mmagic.models.editors.arcface.model_irse.IR_152(input_size)

Constructs a ir-152 model.

mmagic.models.editors.arcface.model_irse.IR_SE_50(input_size)

Constructs a ir_se-50 model.

mmagic.models.editors.arcface.model_irse.IR_SE_101(input_size)

Constructs a ir_se-101 model.

mmagic.models.editors.arcface.model_irse.IR_SE_152(input_size)

Constructs a ir_se-152 model.