[pytorch] ♟작은 고추가 맵다 : MobileNetv2

모델은 작지만, 성능은 강력한 Mobile용 DNN MoblieNetv2를 소개합니다!

https://arxiv.org/abs/1801.04381
https://github.com/tonylins/pytorch-mobilenet-v2

MobileNetv2의 특징

Depthwise Separable Convolution

• channel별 convolution 수행 후, 이후 pointwise 추가 연산
• 연산량 감소 효과

Linear Bottlneck

• narrow - wide - narrow에서 block간 연결에서 ReLU를 사용하지 않음 (ReLU6와는 다름)
• 더 중요한 것은 narrow layer를 중요 정보가 저장되어 있는 것으로 생각, 이를 skip conn.에 태움
• skip conn.을 narrow로 태우기에 연산량 감소 효과

Inverted Residual

• 기존 Resnet과 다르게 narrow - wide - narrow
• ReLU6

코드 분석

"""
Creates a MobileNetV2 Model as defined in:
Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, Liang-Chieh Chen. (2018). 
MobileNetV2: Inverted Residuals and Linear Bottlenecks
arXiv preprint arXiv:1801.04381.
import from https://github.com/tonylins/pytorch-mobilenet-v2
"""

import torch.nn as nn
import math

__all__ = ['mobilenetv2']

# 나뉠수 있게 해주기 => 모든 channel number를 8로 나눌 수 있게 만드러주기! : divisor가 아마 8일듯?
def _make_divisible(v, divisor, min_value=None):
    """
    This function is taken from the original tf repo.
    It ensures that all layers have a channel number that is divisible by 8
    It can be seen here:
    https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet/mobilenet.py
    :param v:
    :param divisor:
    :param min_value:
    :return:
    """
    if min_value is None:
        min_value = divisor
    new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
    # Make sure that round down does not go down by more than 10%.
    if new_v < 0.9 * v:
        new_v += divisor
    return new_v

# conv 33 bn relu6
def conv_3x3_bn(inp, oup, stride):
    return nn.Sequential(
        nn.Conv2d(inp, oup, 3, stride, 1, bias=False),
        nn.BatchNorm2d(oup),
        nn.ReLU6(inplace=True)
    )

# conv11 bn relu6
def conv_1x1_bn(inp, oup):
    return nn.Sequential(
        nn.Conv2d(inp, oup, 1, 1, 0, bias=False),
        nn.BatchNorm2d(oup),
        nn.ReLU6(inplace=True)
    )

# narrow - wide - narrow 형식 InvertedResidual
class InvertedResidual(nn.Module):
    def __init__(self, inp, oup, stride, expand_ratio):
        super(InvertedResidual, self).__init__()
        # stride는 반드시 1, 2여야 함
        assert stride in [1, 2]

        # expand ratio를 이용해서 channel 확장! : narrow - wide - narrow
        hidden_dim = round(inp * expand_ratio)
        # skip connection을 이용하기 위한 w h 크기 check
        self.identity = stride == 1 and inp == oup

        # channel 수 변경이 없는경우? =>  
        if expand_ratio == 1:
            self.conv = nn.Sequential(
                # depthwise
                nn.Conv2d(hidden_dim, hidden_dim, 3, stride, 1, groups=hidden_dim, bias=False), # channel수 == groups => depthwise(channelwise) conv. => 어짜피 그 이후 값 더해주니까 depthwise + pointwise
                nn.BatchNorm2d(hidden_dim),
                nn.ReLU6(inplace=True),
                # pointwise-linear
                nn.Conv2d(hidden_dim, oup, 1, 1, 0, bias=False),
                nn.BatchNorm2d(oup),
            )
        # inverted residual conv
        else:
            self.conv = nn.Sequential(
                # pointwise
                nn.Conv2d(inp, hidden_dim, 1, 1, 0, bias=False),
                nn.BatchNorm2d(hidden_dim),
                nn.ReLU6(inplace=True),
                # depthwise
                nn.Conv2d(hidden_dim, hidden_dim, 3, stride, 1, groups=hidden_dim, bias=False),
                nn.BatchNorm2d(hidden_dim),
                nn.ReLU6(inplace=True),
                # pointwise-linear
                nn.Conv2d(hidden_dim, oup, 1, 1, 0, bias=False),
                nn.BatchNorm2d(oup),
            )

    # skip or not.
    def forward(self, x):
        if self.identity:
            return x + self.conv(x)
        else:
            return self.conv(x)


class MobileNetV2(nn.Module):
    def __init__(self, num_classes=1000, width_mult=1.):
        super(MobileNetV2, self).__init__()
        # setting of inverted residual blocks
        self.cfgs = [
            # t, c, n, s => expansion factor, output channel 수, 반복수, stride
            [1,  16, 1, 1],
            [6,  24, 2, 2],
            [6,  32, 3, 2],
            [6,  64, 4, 2],
            [6,  96, 3, 1],
            [6, 160, 3, 2],
            [6, 320, 1, 1],
        ]

        # building first layer
        # channel 생성 : 4 / 8로 나눠떨어지게
        input_channel = _make_divisible(32 * width_mult, 4 if width_mult == 0.1 else 8)
        layers = [conv_3x3_bn(3, input_channel, 2)] # w h는 half, channel수 => input_channel
        # building inverted residual blocks
        block = InvertedResidual
        for t, c, n, s in self.cfgs:
            output_channel = _make_divisible(c * width_mult, 4 if width_mult == 0.1 else 8)
            for i in range(n):
                layers.append(block(input_channel, output_channel, s if i == 0 else 1, t)) # block 당 stride는 첫번째에만 수행 : 나머지는 1 => 각 block 첫 layer에서만 downsampling
                input_channel = output_channel
        self.features = nn.Sequential(*layers) # 싹다모아서 만들기
        # building last several layers
        output_channel = _make_divisible(1280 * width_mult, 4 if width_mult == 0.1 else 8) if width_mult > 1.0 else 1280
        self.conv = conv_1x1_bn(input_channel, output_channel)
        self.avgpool = nn.AdaptiveAvgPool2d((1, 1)) # channel수로 만들어주고
        self.classifier = nn.Linear(output_channel, num_classes) # FC!

        self._initialize_weights()

    def forward(self, x):
        x = self.features(x)
        x = self.conv(x)
        x = self.avgpool(x)
        x = x.view(x.size(0), -1)
        x = self.classifier(x)
        return x

    def _initialize_weights(self):
        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
                m.weight.data.normal_(0, math.sqrt(2. / n))
                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):
                m.weight.data.normal_(0, 0.01)
                m.bias.data.zero_()

def mobilenetv2(**kwargs):
    """
    Constructs a MobileNet V2 model
    """
    return MobileNetV2(**kwargs)