mobilenetv3代码复现pytorch

### 如何在 PyTorch 中实现 MobileNetV3 网络结构 为了构建 MobileNetV3，在 PyTorch 中可以定义一系列模块来表示网络的不同部分。下面是一个简化版的 `MobileNetV3` 实现，它基于已有的研究和开源项目[^1]。 #### 定义基础组件 首先定义一些必要的组件，比如卷积层、批量归一化以及激活函数 h-swish 和 ReLU： ```python import torch.nn as nn import torch class Hswish(nn.Module): def __init__(self, inplace=True): super(Hswish, self).__init__() self.inplace = inplace def forward(self, x): return x * F.relu6(x + 3., inplace=self.inplace) / 6. def conv_3x3_bn(inp, oup, stride): return nn.Sequential( nn.Conv2d(inp, oup, 3, stride, 1, bias=False), nn.BatchNorm2d(oup), Hswish() ) ``` #### 构建瓶颈 (Bottleneck) 接着创建一个类用于生成每个阶段内的瓶颈单元（bneck），这是 MobileNetV3 的核心组成部分之一: ```python class Bottleneck(nn.Module): def __init__(self, inp, oup, kernel, exp_size, se, nl, s): super(Bottleneck, self).__init__() assert s in [1, 2] hidden_dim = exp_size layers = [] # pw if inp != hidden_dim: layers.append(nn.Conv2d(inp, hidden_dim, 1, 1, 0, bias=False)) layers.append(nn.BatchNorm2d(hidden_dim)) if nl == "RE": layers.append(nn.ReLU(inplace=True)) elif nl == "HS": layers.append(Hswish()) layers.extend([ # dw nn.Conv2d(hidden_dim, hidden_dim, kernel, s, kernel//2, groups=hidden_dim, bias=False), nn.BatchNorm2d(hidden_dim), # SE SELayer(hidden_dim) if se else nn.Identity(), # pw-linear nn.Conv2d(hidden_dim, oup, 1, 1, 0, bias=False), nn.BatchNorm2d(oup), ]) if nl == "RE": layers.append(nn.ReLU(inplace=True)) elif nl == "HS": layers.append(Hswish()) self.conv = nn.Sequential(*layers) self.use_res_connect = s == 1 and inp == oup def forward(self, x): out = self.conv(x) if self.use_res_connect: return x + out else: return out ``` 这里还涉及到 Squeeze-and-Excitation(SE) 层的设计，这有助于提高模型性能[^4]: ```python class SELayer(nn.Module): def __init__(self, channel, reduction=4): super(SELayer, self).__init__() self.avg_pool = nn.AdaptiveAvgPool2d(1) self.fc = nn.Sequential( nn.Linear(channel, channel // reduction, bias=False), nn.ReLU(inplace=True), nn.Linear(channel // reduction, channel, bias=False), Hsigmoid() # or use sigmoid activation function ) def forward(self, x): b, c, _, _ = x.size() y = self.avg_pool(x).view(b, c) y = self.fc(y).view(b, c, 1, 1) return x * y.expand_as(x) ``` 最后组合上述所有部件完成整个 MobileNetV3 结构的搭建，并提供预训练权重加载功能以便快速上手使用[^2]。 ```python cfgs_large = [ # k, t, c, SE, NL, s [3, 16, 16, False, 'RE', 1], [3, 64, 24, False, 'RE', 2], ... ] class MobileNetV3_Large(nn.Module): def __init__(num_classes=1000): super(MobileNetV3_Large, self).__init__() self.features = make_layers(cfgs_large) self.classifier = nn.Sequential( nn.Linear(last_channel, num_classes), ) def forward(self, x): x = self.features(x) x = x.mean([2, 3]) # global average pooling x = self.classifier(x) return x def make_layers(cfgs): layers = [conv_3x3_bn(3, cfgs[0][1], 2)] for k, exp_size, c, use_se, nl, s in cfgs: output_channels = c layers.append(Bottleneck(input_channel, output_channels, k, exp_size, use_se, nl, s)) input_channel = output_channels return nn.Sequential(*layers) model = MobileNetV3_Large().to(device) pretrained_dict = torch.load('path_to_pretrained_weights.pth') model.load_state_dict(pretrained_dict) ``` 以上就是利用 PyTorch 来实现 MobileNetV3 大型版本的一个简单例子。实际应用中可能还需要考虑更多细节优化等问题。