Question

我一直在尝试重新训练模型，但不幸的是，最近两天我一直遇到相同的错误。

您能帮忙一点吗？

初始工作：

%matplotlib inline
%config InlineBackend.figure_format = 'retina'

import matplotlib.pyplot as plt
import numpy as np
import time

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

数据集：

data_dir = 'flowers'
train_dir = data_dir + '/train'

data_dir = 'flowers'

train_transforms = transforms.Compose([transforms.Resize(224),
                                       transforms.RandomResizedCrop(224),
                                       transforms.RandomRotation(45),
                                       transforms.RandomHorizontalFlip(),
                                       transforms.ToTensor(),
                                       transforms.Normalize([0.485, 0.456, 0.406], 
                                                            [0.229, 0.224, 0.225])])


train_data = datasets.ImageFolder(train_dir, transform=train_transforms)

trainloader = torch.utils.data.DataLoader(train_data, batch_size=32, shuffle=True)

import json

with open('cat_to_name.json', 'r') as f:
    cat_to_name = json.load(f)

试图使用预先训练的模型并仅训练分类器：

# Load a pretrained model
model = models.vgg16(pretrained=True)

# Keep the parameters the same
for param in model.parameters():
    param.requires_grad = False


# and final output 102, since tht we have 102 flowers. 
classifier = nn.Sequential(OrderedDict([            
                          ('fc1', nn.Linear(25088, 4096)), 
                          ('relu', nn.ReLU()),
                          ('fc3', nn.Linear(4096, 102)),
                          ('output', nn.LogSoftmax(dim=1))
                          ]))

# Replace model's old classifier with the new classifier
model.classifier = classifier

# Calculate the loss
criterion = nn.NLLLoss()
optimizer = optim.Adam(model.classifier.parameters(), lr=0.001)

model.to('cuda')

epochs = 1
print_every = 40
steps = 0

for e in range(epochs):
    running_loss = 0
    model.train()
   # model = model.double()
    for images, labels in iter(trainloader):
        steps += 1

        images.resize_(32, 3, 224, 224)          

        inputs = Variable(images.to('cuda'))
        targets = Variable(labels.to('cuda'))

        optimizer.zero_grad()

        # Forward and backward passes
        output = model.forward(images)
        loss = criterion(output, labels)
        loss.backward()
        optimizer.step()

        #running_loss += loss.data[0]
        running_loss += loss.item()

        if steps % print_every == 0:
            print("Epoch: {}/{}... ".format(e+1, epochs),
                  "Loss: {:.4f}".format(running_loss/print_every))

错误消息：

RuntimeError：类型为torch.FloatTensor的预期对象，但发现参数＃2 torch.cuda.DoubleTensor的类型为weight

Answer 1

如果要在pyTorch中使用gpu，则必须确保两者 操作流程（这是您的模型）和数据已传输到cuda设备。
通过这个（ 1 ），我的意思是，整个工作流程包括pre-trained model， criterion ， classifier 和inputs应全部分配给cuda设备。
如果model.cuda()不能仅保证（2），则可能有必要对所有对象进行手动操作，以确保权重和输入数据均为cuda float类型。

希望有帮助。

Answer 2

您应该已经将输入传递到前馈网络，但是您已经将图像传递到网络

void main() {
    DivideBy *d = new DivideBy(2);
    Truncate *t = new Truncate();
    Compose<DivideBy, Truncate> *c1 = new Compose<DivideBy,Truncate>(d,t);
    Compose<Truncate, DivideBy> *c2 = new Compose<Truncate, DivideBy>(t,d);
    cout << (*c1)(100.7) << endl; // Prints 50.0 
    cout << (*c2)(11) << endl; // Prints 5
}

RuntimeError：类型为torch.FloatTensor的预期对象，但为参数＃2'weight'

2 个答案: