轻量化模型之mobilenet v3

MobileNet v3

onnx 导出参考：torchvision onnx 模型导出_星魂非梦的博客-CSDN博客

1. 模型描述 

MobileNet v3 来自论文：“Searching for MobileNetV3”.

 2.  onxx 导出

参考：轻量化模型之mobilenet v2_星魂非梦的博客-CSDN博客

 由于 Onnx support hardswish in opset-14 version.  所以opset设为14，需要修改下导出代码：

export_onnx(model, im, file, 14, train = True, dynamic = False, simplify=True)  # opset 14

MobileNet v3 中使用了3种激活函数，分别为：Relu、HardSwish和HardSigmoid(SE模块中使用)。

Hardsigmoid — PyTorch 1.11.0 documentation

Hardswish — PyTorch 1.11.0 documentation

torchvision 中有两个 MobileNet v3 模型我们先介绍 MobileNet V3 Small。

3. MobileNet V3 Small

 上图中，虚线部分表示该模块可选，见上图右表。上表中的 3×3, 5×5 为benck中CBA(dw)的卷积核大小。 

论文中：

补充：

python partial函数

functools.partial(func[,*args][, **kwargs])

第一个参数是函数，后面的是参数。这在SE 模块中被使用：

se_layer: Callable[..., nn.Module] = partial(SElayer, scale_activation=nn.Hardsigmoid)

 SElayer 代码如下：

class SqueezeExcitation(torch.nn.Module):
    def __init__(
        self,
        input_channels: int,
        squeeze_channels: int,
        activation: Callable[..., torch.nn.Module] = torch.nn.ReLU,
        scale_activation: Callable[..., torch.nn.Module] = torch.nn.Sigmoid,
    ) -> None:
        super().__init__()
        self.avgpool = torch.nn.AdaptiveAvgPool2d(1)
        self.fc1 = torch.nn.Conv2d(input_channels, squeeze_channels, 1)
        self.fc2 = torch.nn.Conv2d(squeeze_channels, input_channels, 1)
        self.activation = activation()
        self.scale_activation = scale_activation()

    def _scale(self, input: Tensor) -> Tensor:
        scale = self.avgpool(input)
        scale = self.fc1(scale)
        scale = self.activation(scale)
        scale = self.fc2(scale)
        return self.scale_activation(scale)

    def forward(self, input: Tensor) -> Tensor:
        scale = self._scale(input)
        return scale * input

上图中的benck 模块代码如下：

class InvertedResidual(nn.Module):
    # Implemented as described at section 5 of MobileNetV3 paper
    def __init__(self, cnf: InvertedResidualConfig, norm_layer: Callable[..., nn.Module],
                 se_layer: Callable[..., nn.Module] = partial(SElayer, scale_activation=nn.Hardsigmoid)):
        super().__init__()
        if not (1 <= cnf.stride <= 2):
            raise ValueError('illegal stride value')

        self.use_res_connect = cnf.stride == 1 and cnf.input_channels == cnf.out_channels

        layers: List[nn.Module] = []
        activation_layer = nn.Hardswish if cnf.use_hs else nn.ReLU

        # expand
        if cnf.expanded_channels != cnf.input_channels:
            layers.append(ConvNormActivation(cnf.input_channels, cnf.expanded_channels, kernel_size=1,
                                             norm_layer=norm_layer, activation_layer=activation_layer))

        # depthwise
        stride = 1 if cnf.dilation > 1 else cnf.stride
        layers.append(ConvNormActivation(cnf.expanded_channels, cnf.expanded_channels, kernel_size=cnf.kernel,
                                         stride=stride, dilation=cnf.dilation, groups=cnf.expanded_channels,
                                         norm_layer=norm_layer, activation_layer=activation_layer))
        if cnf.use_se:
            squeeze_channels = _make_divisible(cnf.expanded_channels // 4, 8)
            layers.append(se_layer(cnf.expanded_channels, squeeze_channels))

        # project
        layers.append(ConvNormActivation(cnf.expanded_channels, cnf.out_channels, kernel_size=1, norm_layer=norm_layer,
                                         activation_layer=None))

        self.block = nn.Sequential(*layers)
        self.out_channels = cnf.out_channels
        self._is_cn = cnf.stride > 1

    def forward(self, input: Tensor) -> Tensor:
        result = self.block(input)
        if self.use_res_connect:
            result += input
        return result

4. MobileNet V3 Large

5. 核心模块

InvertedResidual：

 上图为选取的一个代表模块，其中虚线部分可能不存在。具体看上面图。

此外，我自己导出的 onnx 文件链接：

https://download.csdn.net/download/hymn1993/85337590