Pytorch迁移学习之猫狗分类

1.AttributeError: ‘VGG’ object has no attribute 'fc’错误
解决：fc换成classifier[6]，并且带【索引】，索引值可以直接打印模型看到结构层
model= models.vgg16(pretrained=True) for param in model.parameters(): #params have requires_grad=True by default param.requires_grad = False num_ftrs = model.fc.in_features

2.element 0 of tensors does not require grad and does not have a grad_fn 错误
解决：这是数据出现错误，需要设置批数据的requires_grad值 data=Variable(data,requires_grad=True)，标签值y则不用

        for i,(data,y) in enumerate(train_loader):
            data=Variable(data,requires_grad=True)
            # y=Variable(y,requires_grad=True)
            print(data.requires_grad)

数据结构，
3.vgg模型代码

import torch.nn.functional as F
import torch
import torch.nn as nn
from torch.autograd import Variable
import torch.optim as optim
import numpy as np
import torchvision
from torchvision import datasets, models, transforms

data_transform = transforms.Compose([
    transforms.Resize(224),
    transforms.CenterCrop(224),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
])

train_dataset = datasets.ImageFolder(root=r'C:\Users\Lavector\Desktop\项目\分类', transform=data_transform)
test_dataset = datasets.ImageFolder(root=r'C:\Users\Lavector\Desktop\项目\分类test', transform=data_transform)

train_loader = torch.utils.data.DataLoader(train_dataset , batch_size=8, shuffle=True,num_workers=1)
test_loader = torch.utils.data.DataLoader(test_dataset , batch_size=3,shuffle=False,num_workers=1)

model= models.vgg16(pretrained=True)
for param in model.parameters(): #params have requires_grad=True by default
        param.requires_grad = False
# num_ftrs = model.fc.in_features
num_ftrs = model.classifier[6].in_features
model.classifier[6] = nn.Linear(num_ftrs, 2)

print(model)
print(num_ftrs)

cost = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(),lr=0.001)

if __name__ == '__main__':

    model.train(True)
    for epoch in range(10):
        print("===epoc===%d"%epoch)
        for i,(data,y) in enumerate(train_loader):
            data=Variable(data,requires_grad=True)
            # y=Variable(y,requires_grad=True)
            print(data.requires_grad)
            out=model(data)
            print(out)
            loss=cost(out,y)
            optimizer.zero_grad()
            loss.backward()
            optimizer.step()
            print('loss:',loss,loss.item())

    model.train(False)
    # yy=net.test_modle(data)
    # print(yy)
    testacc = 0
    for datas in test_loader:
        images, labels = datas
        images = Variable(images)
        # labels = Variable(labels)
        # print(net.test_modle(images))
        torch.no_grad()
        print(labels)
        outputs=model(images)
        print(outputs)
        print(outputs.size())
        print('***')
        predict = F.softmax(outputs)
        print(predict)
        print("%.2f" % torch.max(predict).item())
        print(torch.max(predict, 1)[0].data.numpy().tolist())  ##返回每行中最大的数，并返回其索引值
        print(torch.max(predict, 1)[1].data.numpy().tolist())

4.resnet模型代码

import torch.nn.functional as F
import torch
import torch.nn as nn
from torch.autograd import Variable
import torch.optim as optim
import numpy as np
import torchvision
from torchvision import datasets, models, transforms

data_transform = transforms.Compose([
    transforms.Resize(224),
    transforms.CenterCrop(224),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])
])

train_dataset = datasets.ImageFolder(root=r'C:\Users\Lavector\Desktop\项目\分类', transform=data_transform)
test_dataset = datasets.ImageFolder(root=r'C:\Users\Lavector\Desktop\项目\分类test', transform=data_transform)

train_loader = torch.utils.data.DataLoader(train_dataset , batch_size=8, shuffle=True,num_workers=1)
test_loader = torch.utils.data.DataLoader(test_dataset , batch_size=3,shuffle=False,num_workers=1)

model= models.resnet34(pretrained=True)
print(model)
# model= models.vgg16(pretrained=True)
for param in model.parameters(): #params have requires_grad=True by default
        param.requires_grad = False
# num_ftrs = model.fc.in_features
num_ftrs = model.fc.in_features
model.fc = nn.Linear(num_ftrs, 2)
# num_ftrs = model.classifier[6].in_features
# model.classifier[6] = nn.Linear(num_ftrs, 2)

print(model)
print(num_ftrs)

cost = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(),lr=0.001)

if __name__ == '__main__':

    model.train(True)
    for epoch in range(10):
        print("===epoc===%d"%epoch)
        for i,(data,y) in enumerate(train_loader):
            data=Variable(data,requires_grad=True)
            # y=Variable(y,requires_grad=True)
            print(data.requires_grad)
            out=model(data)
            print(out)
            loss=cost(out,y)
            optimizer.zero_grad()
            loss.backward()
            optimizer.step()
            print('loss:',loss,loss.item())

    model.train(False)
    # yy=net.test_modle(data)
    # print(yy)
    testacc = 0
    for datas in test_loader:
        images, labels = datas
        images = Variable(images)
        # labels = Variable(labels)
        # print(net.test_modle(images))
        model.eval()
        with torch.no_grad():
            print(labels)
            outputs=model(images)
            print(outputs)
            print(outputs.size())
            print('***')
            predict = F.softmax(outputs)
            print(predict)
            print("%.2f" % torch.max(predict).item())
            print(torch.max(predict, 1)[0].data.numpy().tolist())  ##返回每行中最大的数，并返回其索引值
            print(torch.max(predict, 1)[1].data.numpy().tolist())
            # mm, prediction = torch.max(outputs.data, 1)
            # print(outputs)
            # print(mm)
            # print(prediction)
        #     print(labels)
        #     print(labels.data)
        #
            # testacc += torch.sum(prediction == labels.data)
        #     print(testacc)
        # print(testacc)