Assertion `THIndexTensor_(size)(target, 0) == batch_size' failed

1

I know this has been asked but I could not apply the suggestions there to solve my problem. My code is a simple copy of Pytorch’s tutorial to classify CIFAR-10 dataset and looks like this:

from torch.autograd import Variable
import torch.nn as nn
import torch.nn.functional as F

class Net(nn.Module):
def init(self):
super(Net, self).init()
self.conv1 = nn.Conv2d(3, 32, 3)
self.pool = nn.MaxPool2d(2, 2)
self.conv2 = nn.Conv2d(32, 64, 3)
self.conv3 = nn.Conv2d(64, 128, 3)
self.conv4 = nn.Conv2d(128, 256, 3)
self.conv5 = nn.Conv2d(256, 512, 3)
self.conv6 = nn.Conv2d(512, 1024, 3)
self.fc1 = nn.Linear(36864, 120)
self.fc2 = nn.Linear(120, 84)
self.fc3 = nn.Linear(84, 10)

def forward(self, x):
    #x = self.pool(F.relu(self.conv1(x)))
    x = F.relu(self.conv1(x))
    #x = self.pool(F.relu(self.conv2(x)))
    x = F.relu(self.conv2(x))
    x = self.pool(F.relu(self.conv3(x)))
    x = F.relu(self.conv4(x))
    x = F.relu(self.conv5(x))
    x = self.pool(F.relu(self.conv6(x)))        
    x = x.view(-1, 36864)
    x = F.relu(self.fc1(x))
    x = F.relu(self.fc2(x))
    x = self.fc3(x)
    return x
    #print (x.size())

net = Net()

import torch.optim as optim

criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9)

for epoch in range(2): # loop over the dataset multiple times

running_loss = 0.0
for i, data in enumerate(trainloader, 0):
    # get the inputs
    inputs, labels = data

    # wrap them in Variable
    inputs, labels = Variable(inputs), Variable(labels)

    # zero the parameter gradients
    optimizer.zero_grad()

    # forward + backward + optimize
    outputs = net(inputs)
    print(outputs.size())
    print(outputs)
    print(labels)
    print(labels.size())
    loss = criterion(outputs, labels)
    print(output.size())
    print(labels.size())
    loss.backward()
    optimizer.step()

    # print statistics
    running_loss += loss.data[0]
    if i % 2000 == 1999:    # print every 2000 mini-batches
        print('[%d, %5d] loss: %.3f' %
              (epoch + 1, i + 1, running_loss / 2000))
        running_loss = 0.0

print(‘Finished Training’)

and the error is this:

torch.Size([1, 10])
Variable containing:
1.00000e-02 *
-4.6679 -1.3744 0.1477 5.1275 -8.8241 -0.7043 -1.2692 -7.0015 7.4909 -9.7750
[torch.FloatTensor of size 1x10]

Variable containing:
4
1
1
9
[torch.LongTensor of size 4]

torch.Size([4])

RuntimeErrorTraceback (most recent call last)
in ()
18 print(labels)
19 print(labels.size())
—> 20 loss = criterion(outputs, labels)
21 print(output.size())
22 print(labels.size())

/home/jigyasa/anaconda2/lib/python2.7/site-packages/torch/nn/modules/module.pyc in call(self, *input, **kwargs)
222 for hook in self._forward_pre_hooks.values():
223 hook(self, input)
→ 224 result = self.forward(*input, **kwargs)
225 for hook in self._forward_hooks.values():
226 hook_result = hook(self, input, result)

/home/jigyasa/anaconda2/lib/python2.7/site-packages/torch/nn/modules/loss.pyc in forward(self, input, target)
480 _assert_no_grad(target)
481 return F.cross_entropy(input, target, self.weight, self.size_average,
→ 482 self.ignore_index)
483
484

/home/jigyasa/anaconda2/lib/python2.7/site-packages/torch/nn/functional.pyc in cross_entropy(input, target, weight, size_average, ignore_index)
744 True, the loss is averaged over non-ignored targets.
745 “”"
→ 746 return nll_loss(log_softmax(input), target, weight, size_average, ignore_index)
747
748

/home/jigyasa/anaconda2/lib/python2.7/site-packages/torch/nn/functional.pyc in nll_loss(input, target, weight, size_average, ignore_index)
670 dim = input.dim()
671 if dim == 2:
→ 672 return _functions.thnn.NLLLoss.apply(input, target, weight, size_average, ignore_index)
673 elif dim == 4:
674 return _functions.thnn.NLLLoss2d.apply(input, target, weight, size_average, ignore_index)

/home/jigyasa/anaconda2/lib/python2.7/site-packages/torch/nn/_functions/thnn/auto.pyc in forward(ctx, input, target, *args)
45 output = input.new(1)
46 getattr(ctx._backend, update_output.name)(ctx._backend.library_state, input, target,
—> 47 output, *ctx.additional_args)
48 return output
49

RuntimeError: Assertion `THIndexTensor_(size)(target, 0) == batch_size’ failed. at /opt/conda/conda-bld/pytorch_1503966894950/work/torch/lib/THNN/generic/ClassNLLCriterion.c:54

The batch size I’ve taken is 4. But I think the output of print(labels.size()) should be 10 ijstead of 4 because that is the total number of classes.
What am I missing?

2

Your input likely has batch size 1, which is not equal to the label with size 4.

If you have batch size 4, label should be LongTensor of size [4] representing the correct class indices.

3

I have loaded the data like this:

transform = transforms.Compose(
[transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])

trainset = torchvision.datasets.CIFAR10(root=‘./data’, train=True,
download=True, transform=transform)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=4,
shuffle=True, num_workers=2)

testset = torchvision.datasets.CIFAR10(root=‘./data’, train=False,
download=True, transform=transform)
testloader = torch.utils.data.DataLoader(testset, batch_size=4,
shuffle=False, num_workers=2)

classes = (‘plane’, ‘car’, ‘bird’, ‘cat’,
‘deer’, ‘dog’, ‘frog’, ‘horse’, ‘ship’, ‘truck’)

Here, the batch_size argument is 4. Isn’t everything in order then?

4

Hmm then problem might be elsewhere.

Are you sure that each image gives you 36864 values? Is it possible that this is 4x than it really needs to be?