具有 VGG 层的顺序网络

Question

我想拥有一个具有 VGG 网络特性的顺序网络（我想将我的网络传递给另一个函数，它不支持 VGG 对象并支持 nn.sequential）。

我在 VGG 类中添加了函数 getSequentialVersion 方法来获得带有线性层的顺序网络。但是，显然，网络中存在大小不匹配。

'''VGG for CIFAR10. FC layers are removed.
(c) YANG, Wei 
'''
import torch.nn as nn
import torch.utils.model_zoo as model_zoo
import math


__all__ = [
    'VGG','vgg16_bn',
]


model_urls = {
    'vgg16': 'https://download.pytorch.org/models/vgg16-397923af.pth',
}


class VGG(nn.Module):
    def __init__(self, features, num_classes=1000, cfg_type=None, batch_norm=False, **kwargs):
        super(VGG, self).__init__()
        self.features = features
        self.classifier = nn.Linear(512, num_classes)
        self._initialize_weights()
        self.cfg_type = cfg_type
        self.batch_norm = batch_norm

    def forward(self, x):
        x = self.features(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):
                n = m.weight.size(1)
                m.weight.data.normal_(0, 0.01)
                m.bias.data.zero_()
            
    def getSequentialVersion(self):
        return make_layers(cfg[self.cfg_type], batch_norm=self.batch_norm, flag=True)


def make_layers(cfg, batch_norm=False, flag=False):
    layers = []
    in_channels = 3
    for v in cfg:
        if v == 'M':
            layers += [nn.MaxPool2d(kernel_size=2, stride=2)]
        else:
            conv2d = nn.Conv2d(in_channels, v, kernel_size=3, padding=1, bias=False)
            if batch_norm:
                layers += [conv2d, nn.BatchNorm2d(v), nn.ReLU(inplace=True)]
            else:
                layers += [conv2d, nn.ReLU(inplace=True)]
            in_channels = v
    if flag:
        #for Cifar10
        layers += [nn.Linear(512, 10)]
    return nn.Sequential(*layers)


cfg = {
    'A': [64, 'M', 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512, 'M'],
    'B': [64, 64, 'M', 128, 128, 'M', 256, 256, 'M', 512, 512, 'M', 512, 512, 'M'],
    'D': [64, 64, 'M', 128, 128, 'M', 256, 256, 256, 'M', 512, 512, 512, 'M', 512, 512, 512, 'M'],
    'E': [64, 64, 'M', 128, 128, 'M', 256, 256, 256, 256, 'M', 512, 512, 512, 512, 'M', 512, 512, 512, 512, 'M'],
}


def vgg16_bn(**kwargs):
    """VGG 16-layer model (configuration "D") with batch normalization"""
    print("VGG16-bn")
    model = VGG(make_layers(cfg['D'], batch_norm=True), cfg_type='D', batch_norm=True,**kwargs)
    return model

当我调用 summary(net, ( 3, 32, 32))（用于 cifar10）时，我收到不匹配错误。换句话说，主要问题是当我添加这行layers+= [nn.linear(512, 10)]时。

谁能帮帮我？非常感谢。

错误信息：

  File "./main.py", line 284, in <module>
    summary(net, ( 3, 32, 32))
  File "./anaconda3/envs/my_env/lib/python3.8/site-packages/torchsummary/torchsummary.py", line 72, in summary
    model(*x)
  File ".anaconda3/envs/my_env/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
    result = self.forward(*input, **kwargs)
  File "./anaconda3/envs/my_env/lib/python3.8/site-packages/torch/nn/modules/container.py", line 119, in forward
    input = module(input)
  File "./anaconda3/envs/my_env/lib/python3.8/site-packages/torch/nn/modules/module.py", line 889, in _call_impl
    result = self.forward(*input, **kwargs)
  File "./anaconda3/envs/my_env/lib/python3.8/site-packages/torch/nn/modules/linear.py", line 94, in forward
    return F.linear(input, self.weight, self.bias)
  File "./envs/my_env/lib/python3.8/site-packages/torch/nn/functional.py", line 1753, in linear
    return torch._C._nn.linear(input, weight, bias)
RuntimeError: mat1 dim 1 must match mat2 dim 0

附加信息： 这就是我初始化和使用网络的具体方式：

net = vgg16_bn(depth=args.depth,
                  num_classes=num_classes,
                  growthRate=args.growthRate,
                  compressionRate=args.compressionRate,
                  widen_factor=args.widen_factor,
                  dropRate=args.dropRate,
                  base_width=args.base_width,
                  cardinality=args.cardinality).getSequentialVersion()
net = net.to(args.device)    
module_names = ''
if hasattr(net, 'features'): 
    module_names = 'features'
elif hasattr(net, 'children'):
    module_names = 'children'
else:
    print('unknown net modules...')

summary(net, ( 3, 32, 32))

Answer 1

问题很简单。当flag=True（如getSequentialVersion()）时，缺少Flatten 操作。因此，要解决此问题，您需要像这样添加此操作：

if flag:
    # for Cifar10
    layers += [nn.Flatten(), nn.Linear(512, 10)]  # <<< add Flatten before Linear

在forward 调用中，您可以在其视图形式中看到扁平化：

def forward(self, x):
    x = self.features(x)
    x = x.view(x.size(0), -1)  # here, equivalent to torch.flatten(x, 1)
    x = self.classifier(x)
    return x

这就是您将图层转换为Sequential 时缺少的内容。