状況

PytorchのCIFAR-10チュートリアルを参考に，画像を分類するモデルを作成しました．
CIFAR-10チュートリアルはこちらです：
https://pytorch.org/tutorials/beginner/blitz/cifar10_tutorial.html

このモデルに対して，任意の画像を入力し分類させたいのですが，RuntimeError: shape '[-1, 400]' is invalid for input of size 293904というエラーによって失敗してしまいます．解決方法が分かる方がおりましたら，ご教授いただきたいです．

実行コード

チュートリアルのコード
jupyter形式のソースコードを繋げたものを貼っています．上記URLから直接チュートリアルを除いた方が理解しやすいかもしれないです．

python
1import torch
2import torchvision
3import torchvision.transforms as transforms
4import matplotlib.pyplot as plt
5import numpy as np
6import torch.nn as nn
7import torch.nn.functional as F
8import torch.optim as optim
9
10
11transform = transforms.Compose(
12    [transforms.ToTensor(),
13     transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
14
15trainset = torchvision.datasets.CIFAR10(root='./data', train=True,
16                                        download=True, transform=transform)
17trainloader = torch.utils.data.DataLoader(trainset, batch_size=4,
18                                          shuffle=True, num_workers=2)
19
20testset = torchvision.datasets.CIFAR10(root='./data', train=False,
21                                       download=True, transform=transform)
22testloader = torch.utils.data.DataLoader(testset, batch_size=4,
23                                         shuffle=False, num_workers=2)
24
25classes = ('plane', 'car', 'bird', 'cat',
26           'deer', 'dog', 'frog', 'horse', 'ship', 'truck')
27
28# functions to show an image
29
30def imshow(img):
31    img = img / 2 + 0.5     # unnormalize
32    npimg = img.numpy()
33    plt.imshow(np.transpose(npimg, (1, 2, 0)))
34    plt.show()
35
36
37# get some random training images
38dataiter = iter(trainloader)
39images, labels = dataiter.next()
40
41# show images
42imshow(torchvision.utils.make_grid(images))
43# print labels
44print(' '.join('%5s' % classes[labels[j]] for j in range(4)))
45
46class Net(nn.Module):
47    def __init__(self):
48        super(Net, self).__init__()
49        self.conv1 = nn.Conv2d(3, 6, 5)
50        self.pool = nn.MaxPool2d(2, 2)
51        self.conv2 = nn.Conv2d(6, 16, 5)
52        self.fc1 = nn.Linear(16 * 5 * 5, 120)
53        self.fc2 = nn.Linear(120, 84)
54        self.fc3 = nn.Linear(84, 10)
55
56    def forward(self, x):
57        x = self.pool(F.relu(self.conv1(x)))
58        x = self.pool(F.relu(self.conv2(x)))
59        x = x.view(-1, 16 * 5 * 5)
60        # x = x.view(x.size(0), 16*37*37)
61        x = F.relu(self.fc1(x))
62        x = F.relu(self.fc2(x))
63        x = self.fc3(x)
64        return x
65
66net = Net()
67
68criterion = nn.CrossEntropyLoss()
69optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9)
70
71for epoch in range(2):  # loop over the dataset multiple times
72
73    running_loss = 0.0
74    for i, data in enumerate(trainloader, 0):
75        # get the inputs; data is a list of [inputs, labels]
76        inputs, labels = data
77
78        # zero the parameter gradients
79        optimizer.zero_grad()
80
81        # forward + backward + optimize
82        outputs = net(inputs)
83        loss = criterion(outputs, labels)
84        loss.backward()
85        optimizer.step()
86
87        # print statistics
88        running_loss += loss.item()
89        if i % 2000 == 1999:    # print every 2000 mini-batches
90            print('[%d, %5d] loss: %.3f' %
91                  (epoch + 1, i + 1, running_loss / 2000))
92            running_loss = 0.0
93
94print('Finished Training')
95
96PATH = './cifar_net.pth'
97torch.save(net.state_dict(), PATH)
98
99dataiter = iter(testloader)
100images, labels = dataiter.next()
101
102# print images
103imshow(torchvision.utils.make_grid(images))
104print('GroundTruth: ', ' '.join('%5s' % classes[labels[j]] for j in range(4)))
105
106net = Net()
107net.load_state_dict(torch.load(PATH))
108
109_, predicted = torch.max(outputs, 1)
110
111print('Predicted: ', ' '.join('%5s' % classes[predicted[j]]
112                              for j in range(4)))

実行したいコード
調べながら書きました．jpgファイルをTensorに変換して入力させようとしています．

python
1from PIL import Image
2import torchvision.transforms as transforms
3
4device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
5
6transform = transforms.Compose(
7    [transforms.ToTensor(),
8     transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
9 
10def image_loader(image_name):
11    image = Image.open(image_name)
12    image = transform(image).unsqueeze(0)
13    return image.to(device, torch.float)
14 
15img = image_loader("./photo.jpg")
16
17outputs = net(img)