3.字符识别模型
2.1 输入数据信息
dataset加载图片时将图片尺寸固定成了(60, 120)
2.2 加载预训练模型
2.2.1 torchvision.models里有哪些模型?
import torchvision.models as models
Alexnet
- model = models.alexnet(pretrained=False)
VGG
- model = models.vgg11(pretrained =False)
- model = models.vgg11_bn(pretrained =False)
- model = models.vgg13(pretrained =False)
- model = models.vgg13_bn(pretrained =False)
- model = models.vgg16(pretrained =False)
- model = models.vgg16_bn(pretrained =False)
- model = models.vgg19(pretrained =False)
- model = models.vgg19_bn(pretrained =False)
ResNet
- model = models.resnet18(pretrained=False)
- model = models.resnet34(pretrained=False)
- model = models.resnet50(pretrained=False)
- model = models.resnet101(pretrained=False)
- model = models.resnet152(pretrained=False)
SqueezeNet
- model = models.squeezenet1_0(pretrained=False)
- model = models.squeezenet1_1(pretrained=False)
DenseNet
- model = models.densenet121(pretrained=False)
- model = models.densenet169(pretrained=False)
- model = models.densenet161(pretrained=False)
- model = models.densenet201(pretrained=False)
Inception v3
- model = models.inception_v3(pretrained=False)
GoogLeNet
- model = models.googlenet(pretrained=False)
还有几个分类网络,以及许多目标检测网络。
2.2.2 加载预训练模型参数
- 结构一样,加载全部参数
net = model(*args, **kwargs)
net.load_state_dict(torch.load("my_resnet.pth"))
或者
my_resnet = torch.load("my_resnet.pth")
- 加载部分预训练模型
其实大多数时候我们需要根据我们的任务调节我们的模型,所以很难保证模型和公开的模型完全一样,但是预训练模型的参数确实有助于提高训练的准确率,为了结合二者的优点,就需要我们加载部分预训练模型。
pretrained_dict = model_zoo.load_url(model_urls['resnet152'])
model_dict = model.state_dict()
# 将pretrained_dict里不属于model_dict的键剔除掉
pretrained_dict = {k: v for k, v in pretrained_dict.items() if k in model_dict}
# 更新现有的model_dict
model_dict.update(pretrained_dict)
# 加载我们真正需要的state_dict
model.load_state_dict(model_dict)
因为需要剔除原模型中不匹配的键,也就是层的名字,所以我们的新模型改变了的层需要和原模型对应层的名字不一样,比如:resnet最后一层的名字是fc(PyTorch中),那么我们修改过的resnet的最后一层就不能取这个名字,可以叫fc_
3.微改基础模型
对于我们的任务而言有些层并不是直接能用,需要我们微微改一下,比如,resnet最后的全连接层是分1000类,而我们只有21类;又比如,resnet第一层卷积接收的通道是3, 我们可能输入图片的通道是4,那么可以通过以下方法修改:
resnet.conv1 = nn.Conv2d(4, 64, kernel_size=7, stride=2, padding=3, bias=False)
resnet.fc = nn.Linear(2048, 21)
2.2.3 快速加载预训练模型参数
直接使用默认下载权重方式可能会很慢。
- 可以修改源码,使得模型加载已经下载好的参数:
model_urls = { 'resnet50': '/home/huihua/NewDisk1/pretrain_parameter/resnet50-19c8e357.pth',}
- 将模型权重下载至pytorch的缓存文件中:
cd .cache/torch/checkpoints # windows
cd /home/team/.torch/models # linux
2.3 创建自己的CNN模型
class SVHN_Model1(nn.Module):
def __init__(self):
super(SVHN_Model1, self).__init__()
model_conv = models.resnet18(pretrained=True)
model_conv.avgpool = nn.AdaptiveAvgPool2d(1)
model_conv = nn.Sequential(*list(model_conv.children())[:-1])
self.cnn = model_conv
self.fc1 = nn.Linear(512, 11)
self.fc2 = nn.Linear(512, 11)
self.fc3 = nn.Linear(512, 11)
self.fc4 = nn.Linear(512, 11)
self.fc5 = nn.Linear(512, 11)
def forward(self, img):
feat = self.cnn(img)
# print(feat.shape)
feat = feat.view(feat.shape[0], -1)
c1 = self.fc1(feat)
c2 = self.fc2(feat)
c3 = self.fc3(feat)
c4 = self.fc4(feat)
c5 = self.fc5(feat)
return c1, c2, c3, c4, c5