mobilefacenet/mobilefacenet_impl.py

import torch
from torch import nn

##################################  Original Arcface Model #############################################################

class Flatten(nn.Module):
    def forward(self, input):
        return input.view(input.size(0), -1)

##################################  MobileFaceNet #############################################################
    
class Conv_block(nn.Module):
    def __init__(self, in_c, out_c, kernel=(1, 1), stride=(1, 1), padding=(0, 0), groups=1):
        super(Conv_block, self).__init__()
        self.conv = nn.Conv2d(in_c, out_channels=out_c, kernel_size=kernel, groups=groups, stride=stride, padding=padding, bias=False)
        self.bn = nn.BatchNorm2d(out_c)
        self.prelu = nn.PReLU(out_c)
    def forward(self, x):
        x = self.conv(x)
        x = self.bn(x)
        x = self.prelu(x)
        return x

class Linear_block(nn.Module):
    def __init__(self, in_c, out_c, kernel=(1, 1), stride=(1, 1), padding=(0, 0), groups=1):
        super(Linear_block, self).__init__()
        self.conv = nn.Conv2d(in_c, out_channels=out_c, kernel_size=kernel, groups=groups, stride=stride, padding=padding, bias=False)
        self.bn = nn.BatchNorm2d(out_c)
    def forward(self, x):
        x = self.conv(x)
        x = self.bn(x)
        return x

class Depth_Wise(nn.Module):
     def __init__(self, in_c, out_c, residual = False, kernel=(3, 3), stride=(2, 2), padding=(1, 1), groups=1):
        super(Depth_Wise, self).__init__()
        self.conv = Conv_block(in_c, out_c=groups, kernel=(1, 1), padding=(0, 0), stride=(1, 1))
        self.conv_dw = Conv_block(groups, groups, groups=groups, kernel=kernel, padding=padding, stride=stride)
        self.project = Linear_block(groups, out_c, kernel=(1, 1), padding=(0, 0), stride=(1, 1))
        self.residual = residual
     def forward(self, x):
        if self.residual:
            short_cut = x
        x = self.conv(x)
        x = self.conv_dw(x)
        x = self.project(x)
        if self.residual:
            output = short_cut + x
        else:
            output = x
        return output

class Residual(nn.Module):
    def __init__(self, c, num_block, groups, kernel=(3, 3), stride=(1, 1), padding=(1, 1)):
        super(Residual, self).__init__()
        modules = []
        for _ in range(num_block):
            modules.append(Depth_Wise(c, c, residual=True, kernel=kernel, padding=padding, stride=stride, groups=groups))
        self.model = nn.Sequential(*modules)
    def forward(self, x):
        return self.model(x)

class GNAP(nn.Module):
    def __init__(self, embedding_size):
        super(GNAP, self).__init__()
        assert embedding_size == 512
        self.bn1 = nn.BatchNorm2d(512, affine=False)
        self.pool = nn.AdaptiveAvgPool2d((1, 1))

        self.bn2 = nn.BatchNorm1d(512, affine=False)

    def forward(self, x):
        x = self.bn1(x)
        x_norm = torch.norm(x, 2, 1, True)
        x_norm_mean = torch.mean(x_norm)
        weight = x_norm_mean / x_norm
        x = x * weight
        x = self.pool(x)
        x = x.view(x.shape[0], -1)
        feature = self.bn2(x)
        return feature

class GDC(nn.Module):
    def __init__(self, embedding_size):
        super(GDC, self).__init__()
        self.conv_6_dw = Linear_block(512, 512, groups=512, kernel=(7,7), stride=(1, 1), padding=(0, 0))
        self.conv_6_flatten = nn.Flatten()
        self.linear = nn.Linear(512, embedding_size, bias=False)
        #self.bn = BatchNorm1d(embedding_size, affine=False)
        self.bn = nn.BatchNorm1d(embedding_size)

    def forward(self, x):
        x = self.conv_6_dw(x)
        x = self.conv_6_flatten(x)
        x = self.linear(x)
        x = self.bn(x)
        return x

class MobileFaceNet(nn.Module):
    def __init__(self, input_size, embedding_size = 512, output_name = "GDC"):
        super(MobileFaceNet, self).__init__()
        assert output_name in ["GNAP", 'GDC']
        assert input_size[0] in [112]
        self.conv1 = Conv_block(3, 64, kernel=(3, 3), stride=(2, 2), padding=(1, 1))
        self.conv2_dw = Conv_block(64, 64, kernel=(3, 3), stride=(1, 1), padding=(1, 1), groups=64)
        self.conv_23 = Depth_Wise(64, 64, kernel=(3, 3), stride=(2, 2), padding=(1, 1), groups=128)
        self.conv_3 = Residual(64, num_block=4, groups=128, kernel=(3, 3), stride=(1, 1), padding=(1, 1))
        self.conv_34 = Depth_Wise(64, 128, kernel=(3, 3), stride=(2, 2), padding=(1, 1), groups=256)
        self.conv_4 = Residual(128, num_block=6, groups=256, kernel=(3, 3), stride=(1, 1), padding=(1, 1))
        self.conv_45 = Depth_Wise(128, 128, kernel=(3, 3), stride=(2, 2), padding=(1, 1), groups=512)
        self.conv_5 = Residual(128, num_block=2, groups=256, kernel=(3, 3), stride=(1, 1), padding=(1, 1))
        self.conv_6_sep = Conv_block(128, 512, kernel=(1, 1), stride=(1, 1), padding=(0, 0))
        if output_name == "GNAP":
            self.output_layer = GNAP(512)
        else:
            self.output_layer = GDC(embedding_size)

        self._initialize_weights()

    def _initialize_weights(self):
        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
                if m.bias is not None:
                    m.bias.data.zero_()
            elif isinstance(m, nn.BatchNorm2d):
                m.weight.data.fill_(1)
                m.bias.data.zero_()
            elif isinstance(m, nn.Linear):
                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
                if m.bias is not None:
                    m.bias.data.zero_()


    def forward(self, x):
        out = self.conv1(x)

        out = self.conv2_dw(out)

        out = self.conv_23(out)

        out = self.conv_3(out)

        out = self.conv_34(out)

        out = self.conv_4(out)

        out = self.conv_45(out)

        out = self.conv_5(out)

        conv_features = self.conv_6_sep(out)
        out = self.output_layer(conv_features)
        return out, conv_features