RuntimeError: applying transform <monai.transforms.spatial.array.Resize object at 0x7f79268cdc50>

[CS-Rorita]

Hi, I am doing a project with a private data using this notebook https://colab.research.google.com/github/Project-MONAI/tutorials/blob/master/3d_classification/torch/densenet_training_array.ipynb for 3d classification .

When I run the file on the data attached to it, it works fine
While when I only changed the data with the same number of samples but my private data
I had some errors, I searched a lot, but I couldn't find the right solution.
The first epoch started to work and then stopped, and the main error when running the model was as follows:

epoch 1/5
/usr/local/lib/python3.7/dist-packages/monai/data/utils.py:650: UserWarning: Modifying image pixdim from [5.       0.390625 0.390625 1.      ] to [  4.99999992   0.390625     0.39062499 190.99360732]
  warnings.warn(f"Modifying image pixdim from {pixdim} to {norm}")
1/5, train_loss: 0.8953
2/5, train_loss: 0.7255
---------------------------------------------------------------------------
RuntimeError        Caught RuntimeError in DataLoader worker process 0.
Original Traceback (most recent call last):
  File "/usr/local/lib/python3.7/dist-packages/monai/transforms/transform.py", line 89, in apply_transform
    return _apply_transform(transform, data, unpack_items)
  File "/usr/local/lib/python3.7/dist-packages/monai/transforms/transform.py", line 53, in _apply_transform
    return transform(parameters)
  File "/usr/local/lib/python3.7/dist-packages/monai/transforms/spatial/array.py", line 670, in __call__
    "len(spatial_size) must be greater or equal to img spatial dimensions, "
ValueError: len(spatial_size) must be greater or equal to img spatial dimensions, got spatial_size=3 img=4.

this is my code :

images = [
    "/content/asd (1).nii.gz","/content/asd (2).nii.gz","/content/asd (3).nii.gz","/content/asd (4).nii.gz","/content/asd (5).nii.gz",
    "/content/asd (6).nii.gz","/content/asd (7).nii.gz","/content/asd (8).nii.gz","/content/asd (9).nii.gz","/content/healthy (1).nii.gz",
    "/content/healthy (2).nii.gz","/content/healthy (3).nii.gz","/content/healthy (4).nii.gz","/content/healthy (5).nii.gz","/content/healthy (6).nii.gz",
    "/content/healthy (7).nii.gz","/content/healthy (8).nii.gz","/content/healthy (9).nii.gz","/content/healthy (10).nii.gz","/content/healthy (11).nii.gz"
]

# 2 binary labels for gender classification: man or woman
labels = np.array([0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1])

# Represent labels in one-hot format for binary classifier training,
# BCEWithLogitsLoss requires target to have same shape as input
labels = torch.nn.functional.one_hot(torch.as_tensor(labels)).float()

# Define transforms
train_transforms = Compose([ScaleIntensity(), AddChannel(), Resize((96, 96, 96)), RandRotate90(), EnsureType()])

val_transforms = Compose([ScaleIntensity(), AddChannel(), Resize((96, 96, 96)), EnsureType()])

# Define nifti dataset, data loader
check_ds = ImageDataset(image_files=images, labels=labels, transform=train_transforms)
check_loader = DataLoader(check_ds, batch_size=3, num_workers=2, pin_memory=pin_memory)

im, label = monai.utils.misc.first(check_loader)
print(type(im), im.shape, label, label.shape)

# create a training data loader
train_ds = ImageDataset(image_files=images[:10], labels=labels[:10], transform=train_transforms)
train_loader = DataLoader(train_ds, batch_size=2, shuffle=True, num_workers=2, pin_memory=pin_memory)

# create a validation data loader
val_ds = ImageDataset(image_files=images[-10:], labels=labels[-10:], transform=val_transforms)
val_loader = DataLoader(val_ds, batch_size=2, num_workers=2, pin_memory=pin_memory)

# Create DenseNet121, CrossEntropyLoss and Adam optimizer
model = monai.networks.nets.DenseNet121(spatial_dims=3, in_channels=1, out_channels=2).to(device)

loss_function = torch.nn.CrossEntropyLoss()
# loss_function = torch.nn.BCEWithLogitsLoss()  # also works with this data

optimizer = torch.optim.Adam(model.parameters(), 1e-4)

# start a typical PyTorch training
val_interval = 2
best_metric = -1
best_metric_epoch = -1
epoch_loss_values = []
metric_values = []
writer = SummaryWriter()
max_epochs = 5

for epoch in range(max_epochs):
    print("-" * 10)
    print(f"epoch {epoch + 1}/{max_epochs}")
    model.train()
    epoch_loss = 0
    step = 0

    for batch_data in train_loader:
        step += 1
        inputs, labels = batch_data[0].to(device), batch_data[1].to(device)
        optimizer.zero_grad()
        outputs = model(inputs)
        loss = loss_function(outputs, labels)

        loss.backward()
        optimizer.step()
        epoch_loss += loss.item()
        epoch_len = len(train_ds) // train_loader.batch_size
        print(f"{step}/{epoch_len}, train_loss: {loss.item():.4f}")
        writer.add_scalar("train_loss", loss.item(), epoch_len * epoch + step)

    epoch_loss /= step
    epoch_loss_values.append(epoch_loss)
    print(f"epoch {epoch + 1} average loss: {epoch_loss:.4f}")

    if (epoch + 1) % val_interval == 0:
        model.eval()

        num_correct = 0.0
        metric_count = 0
        for val_data in val_loader:
            val_images, val_labels = val_data[0].to(device), val_data[1].to(device)
            with torch.no_grad():
                val_outputs = model(val_images)
                value = torch.eq(val_outputs.argmax(dim=1), val_labels.argmax(dim=1))
                metric_count += len(value)
                num_correct += value.sum().item()

        metric = num_correct / metric_count
        metric_values.append(metric)

        if metric > best_metric:
            best_metric = metric
            best_metric_epoch = epoch + 1
            torch.save(model.state_dict(), "best_metric_model_classification3d_array.pth")
            print("saved new best metric model")

        print(f"Current epoch: {epoch+1} current accuracy: {metric:.4f} ")
        print(f"Best accuracy: {best_metric:.4f} at epoch {best_metric_epoch}")
        writer.add_scalar("val_accuracy", metric, epoch + 1)

print(f"Training completed, best_metric: {best_metric:.4f} at epoch: {best_metric_epoch}")
writer.close()

any help please >
Thanks in advanse

[Nic-Ma]

Hi @CS-Rorita ,

Thanks for your interest here and experiments.
May I know your input data shape?
I suspect it already has the channel dim, so no need the AddChannel transform in the original notebook, you may use AsChannelFirst instead.

Thanks.