Why RuntimeError: Assertion `cur_target >= 0 && cur_target < n_classes' failed

I implemented the code like below:

class Model(nn.Module):
    def __init__(self):
        super(Model, self).__init__()

        self.conv = nn.Sequential(
            nn.Conv2d(3, 64, kernel_size=3, stride=1, padding=1),
            nn.Conv2d(64, 128, kernel_size=3, stride=1, padding=1),

        self.dense = nn.Sequential(
            nn.Linear(16*16*128, 1024),
            nn.Linear(1024, 10)

    def forward(self, input):
        output = self.conv(input)
        output = output.view(-1, 16*16*128)
        output = self.dense(output)

        return output

model = Model()

cost = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters())

n_epochs = 5

for epoch in range(n_epochs):
    running_loss = 0.0
    running_correct = 0
    print("Epoch {}/{}".format(epoch, n_epochs))
    print("-" * 10)
    for data in data_loader_train:
        images, labels = data
        images, labels = Variable(images), Variable(labels)
        outputs = model(images)
        _,pred = torch.max(outputs.data, 1)
        loss = cost(outputs, labels)

        running_loss += loss.data.item()
        running_correct += torch.sum(pred == labels.data)
    testing_correct = 0
    for data in data_loader_test:
        images, labels = data
        images, labels = Variable(images), Variable(labels)
        outputs = model(images)
        _,pred = torch.max(outputs.data, 1)
        testing_correct += torch.sum(pred == labels.data)
    print("Loss is:{:.4f}, Train Accuracy is:{:.4f}%, Test Accuracy is:{:.4f}".format(running_loss/len(data_train), 100*running_correct/len(data_train), 100*testing_correct/len(data_test)))

But I got the error code like this(line 94 is loss = cost(outputs, labels)):

RuntimeError: Assertion `cur_target >= 0 && cur_target < n_classes' failed.  at C:\w\1\s\tmp_conda_3.7_055457\conda\conda-bld\pytorch_1565416617654\work\aten\src\THNN/generic/ClassNLLCriterion.c:94

How can I solve this problem? Could you help me? Thank you!

Could you check the values of your labels?
It should contain class indices in the range [0, nb_classes-1], which is [0, 9],based on your output layer.
Apparently some batches contain label indices outside of this range.

Thanks for your solution! But sorry I’m a newcomer to Pytorch, how can I check the values of labels and make all batches contain label indices within range?

It depends a bit on how you’ve created the Dataset or the targets in particular.
E.g. if you are dealing with image data (images stored in two folders), torchvision.datasets.ImageFolder will read the data and create the appropriate targets.

You could check the range of the labels by printing them in the DataLoader loop, e.g. simply by:


If you could post your Dataset, we could have a look at it here.

Thanks for your help, I check the range of the labels, tensor(2), tensor(98), It looks like far beyond the range.
I implemented my Dataset like below:

train_transformations = transforms.Compose([
    transforms.RandomCrop(32, padding=4),
    transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))

data_train = datasets.CIFAR100(root="./data/", transform=train_transformations, train=True, download=True)

test_transformations = transforms.Compose([
    transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))

data_test = datasets.CIFAR100(root="./data/", transform=test_transformations, train=False)
data_loader_train = torch.utils.data.DataLoader(dataset=data_train, batch_size=64, shuffle=True)
data_loader_test = torch.utils.data.DataLoader(dataset=data_test, batch_size=64, shuffle=True)

Thank you!

CIFAR100 contains images for 100 classes (this is what the 100 stands for), so you need to change the output units to 100 in your last linear layer. If you want to use just 10 classes, you should use CIFAR10 instead.

1 Like

It works! I really appreciate your help! I will try harder to learn Pytorch!!

1 Like

Happy to help and don’t hesitate to ask questions here in case you get stuck somewhere. :wink:

Hi, can you please show code on how to shift label to start with 0 and not 1, mine start with 1?

Sure, you can just subtract 1 from the tensor:

x = torch.randint(1, 10, (20,))
x = x - 1