### 构建ResNeXt50网络
#### 一、理解ResNeXt架构原理
ResNeXt（Residual Next）是在ResNet的基础上进行了改进，通过引入分组卷积的概念来增加网络宽度而不是深度。在ResNet中，输入的具有256个通道的特征经过1×1卷积压缩4倍到64个通道，之后3×3的卷积核用于处理特征，再经1×1卷积扩大通道数与原特征残差连接后输出[^3]。
而在ResNeXt中，采用了多个平行分支的设计思路，在每个分支内执行相同的转换操作，并最终将这些分支的结果相加作为该模块的整体输出。这种设计允许模型自动探索最佳的操作组合方式，从而提高了表达能力而不显著增加计算成本。
#### 二、PyTorch实现ResNeXt50的具体步骤
为了更好地展示如何利用Python中的PyTorch库创建一个完整的ResNeXt50模型，下面给出了一段详细的代码示例：
```python
import torch.nn as nn
class Bottleneck(nn.Module):
expansion = 4
def __init__(self, inplanes, planes, stride=1, downsample=None, cardinality=32):
super(Bottleneck, self).__init__()
# 定义基础路径上的三个连续卷积层
self.conv1 = nn.Conv2d(inplanes, planes * 2, kernel_size=1, bias=False)
self.bn1 = nn.BatchNorm2d(planes * 2)
# 使用cardinality参数控制分组数量
self.conv2 = nn.Conv2d(
planes * 2,
planes * 2,
kernel_size=3,
stride=stride,
padding=1,
groups=cardinality,
bias=False
)
self.bn2 = nn.BatchNorm2d(planes * 2)
self.conv3 = nn.Conv2d(planes * 2, planes * self.expansion, kernel_size=1, bias=False)
self.bn3 = nn.BatchNorm2d(planes * self.expansion)
self.relu = nn.ReLU(inplace=True)
self.downsample = downsample
self.stride = stride
def forward(self, x):
residual = x
out = self.conv1(x)
out = self.bn1(out)
out = self.relu(out)
out = self.conv2(out)
out = self.bn2(out)
out = self.relu(out)
out = self.conv3(out)
out = self.bn3(out)
if self.downsample is not None:
residual = self.downsample(x)
out += residual
out = self.relu(out)
return out
def make_layer(block, inplanes, planes, blocks, stride=1, cardinality=32):
layers = []
# 创建第一个Bottleneck block实例时可能需要调整维度大小
downsample = None
if stride != 1 or inplanes != planes * block.expansion:
downsample = nn.Sequential(
nn.Conv2d(inplanes, planes * block.expansion, kernel_size=1, stride=stride, bias=False),
nn.BatchNorm2d(planes * block.expansion),
)
layers.append(block(inplanes, planes, stride, downsample, cardinality))
# 对于后续的blocks只需要保持相同尺寸即可
inplanes = planes * block.expansion
for i in range(1, blocks):
layers.append(block(inplanes, planes, cardinality=cardinality))
return nn.Sequential(*layers)
class ResNeXt(nn.Module):
def __init__(self, block, layers, num_classes=1000, cardinality=32):
self.cardinality = cardinality
self.inplanes = 64
super(ResNeXt, self).__init__()
# 初始7x7卷积+最大池化
self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,bias=False)
self.bn1 = nn.BatchNorm2d(64)
self.relu = nn.ReLU(inplace=True)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
# 堆叠四个不同尺度下的resnext layer
self.layer1 = make_layer(block, self.inplanes, 64, layers[0], cardinality=self.cardinality)
self.layer2 = make_layer(block, self.inplanes*block.expansion, 128, layers[1], stride=2, cardinality=self.cardinality)
self.layer3 = make_layer(block, self.inplanes*(block.expansion**2), 256, layers[2], stride=2, cardinality=self.cardinality)
self.layer4 = make_layer(block, self.inplanes*(block.expansion**3), 512, layers[3], stride=2, cardinality=self.cardinality)
self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
self.fc = nn.Linear(512 * block.expansion, num_classes)
# 初始化权重
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
def _forward_impl(self, x):
x = self.conv1(x)
x = self.bn1(x)
x = self.relu(x)
x = self.maxpool(x)
x = self.layer1(x)
x = self.layer2(x)
x = self.layer3(x)
x = self.layer4(x)
x = self.avgpool(x)
x = torch.flatten(x, 1)
x = self.fc(x)
return x
def forward(self, x):
return self._forward_impl(x)
def resnext50_32x4d(pretrained=False, progress=True, **kwargs):
model = ResNeXt(Bottleneck, [3, 4, 6, 3], cardinality=32, **kwargs)
if pretrained:
state_dict = load_state_dict_from_url(model_urls['resnext50_32x4d'],progress=progress)
model.load_state_dict(state_dict)
return model
```
这段代码定义了一个名为`ResNeXt` 的类，其中包含了构建整个ResNeXt50所需的主要组件——瓶颈结构(`Bottleneck`)和堆叠函数(`make_layer`)。此外还实现了预训练版本加载的功能以便快速上手使用已经过良好调优过的初始权值设置。
超级会员免费看
订阅专栏 解锁全文
扫一扫
606
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
