MobileNetsV2代码复现

### MobileNetV2 的代码实现与教程 #### 使用 PyTorch 实现 MobileNetV2 以下是基于 PyTorch 的 MobileNetV2 完整实现： ```python import torch.nn as nn import torch class ConvBNReLU(nn.Sequential): def __init__(conv, in_planes, out_planes, kernel_size=3, stride=1, groups=1): padding = (kernel_size - 1) // 2 super(ConvBNReLU, conv).__init__( nn.Conv2d(in_planes, out_planes, kernel_size, stride, padding, groups=groups, bias=False), nn.BatchNorm2d(out_planes), nn.ReLU6(inplace=True) ) class InvertedResidual(nn.Module): def __init__(self, inp, oup, stride, expand_ratio): super(InvertedResidual, self).__init__() hidden_dim = int(round(inp * expand_ratio)) self.use_res_connect = stride == 1 and inp == oup layers = [] if expand_ratio != 1: # pw layers.append(ConvBNReLU(inp, hidden_dim, kernel_size=1)) layers.extend([ # dw ConvBNReLU(hidden_dim, hidden_dim, stride=stride, groups=hidden_dim), # pw-linear nn.Conv2d(hidden_dim, oup, 1, 1, 0, bias=False), nn.BatchNorm2d(oup), ]) conv.block = nn.Sequential(*layers) def forward(self, x): if conv.use_res_connect: return x + conv.block(x) else: return conv.block(x) class MobileNetV2(nn.Module): def __init__(mobilenetv2, num_classes=1000, width_mult=1.0, inverted_residual_setting=None, round_nearest=8, block=None): last_channel = 1280 input_channel = 32 cfgs = [ # t, c, n, s [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], ] if block is None: block = InvertedResidual mobilenetv2.features = [] # building first layer input_channel = _make_divisible(input_channel * width_mult, round_nearest) last_channel = _make_divisible(last_channel * max(1.0, width_mult), round_nearest) mobilenetv2.features.append(ConvBNReLU(3, input_channel, stride=2)) # building inverted residual blocks for t, c, n, s in cfgs: output_channel = _make_divisible(c * width_mult, round_nearest) for i in range(n): stride = s if i == 0 else 1 mobilenetv2.features.append(block(input_channel, output_channel, stride, expand_ratio=t)) input_channel = output_channel # building last several layers mobilenetv2.features.append(ConvBNReLU(input_channel, last_channel, kernel_size=1)) # make it nn.Sequential mobilenetv2.features = nn.Sequential(*mobilenetv2.features) # building classifier mobilenetv2.classifier = nn.Sequential( nn.Dropout(0.2), nn.Linear(last_channel, num_classes), ) def forward(mobilenetv2, x): x = mobilenetv2.features(x) x = x.mean([2, 3]) # global average pooling x = mobilenetv2.classifier(x) return x ``` 这段代码展示了如何定义 `ConvBNReLU` 和 `InvertedResidual` 层以及完整的 `MobileNetV2` 类。此网络结构采用了倒残差模块和线性瓶颈设计，这些特性使得模型更加高效[^1]。 对于 TensorFlow 用户来说，官方提供了预训练好的 MobileNetV2 模型可以直接加载并使用: ```python from tensorflow.keras.applications import MobileNetV2 model = MobileNetV2(weights='imagenet') ``` 这行简单的命令即可获取到已经训练好 ImageNet 数据集的 MobileNetV2 模型实例[^2]。