[Bug]: Training fails with dataloader

[stevejonathancitriot]

Describe the bug

I get an error when I try to access the dataloader items with the hazelnut_toy dataset

Dataset

Folder

Model

PatchCore

Steps to reproduce the behavior

Installed anomalib with pip and "anomalib install"

Error when running this code

from anomalib.data import Folder

Create the datamodule

datamodule = Folder(
name="hazelnut_toy",
root="/home/citriot/Downloads/hazelnut_toy",
normal_dir="good",
abnormal_dir="crack",
mask_dir="mask/crack",
normal_split_ratio=0.2,
)

Setup the datamodule

datamodule.setup()

train_dataset = datamodule.train_data
val_dataset = datamodule.val_data
test_dataset = datamodule.test_data

Access the dataloaders

train_dataloader = datamodule.train_dataloader()
val_dataloader = datamodule.val_dataloader()
test_dataloader = datamodule.test_dataloader()

i, train_data = next(enumerate(datamodule.train_dataloader()))
print(train_data.keys(), train_data['image'].shape)
# dict_keys(['image_path', 'label', 'image', 'mask_path', 'mask'])

i, val_data = next(enumerate(datamodule.val_dataloader()))
print(val_data.keys(), val_data['image'].shape)
# dict_keys(['image_path', 'label', 'image', 'mask_path', 'mask'])

i, test_data = next(enumerate(datamodule.test_dataloader()))
print(test_data.keys(), test_data['image'].shape)
# dict_keys(['image_path', 'label', 'image', 'mask_path', 'mask'])

from anomalib.models import Patchcore
from anomalib.engine import Engine

Create the model and engine

model = Patchcore()
engine = Engine()

Train a Patchcore model on the given datamodule

engine.train(datamodule=datamodule, model=model)

OS information

OS information:

• OS: [e.g. Ubuntu 22.04]
• Python version: [e.g. 3.10.0]
• Anomalib version: [e.g. 1.0.0]
• PyTorch version: [e.g. 2.1.2+cu121]
• CUDA/cuDNN version: [e.g. 12.1]
• GPU models and configuration: [e.g. 1x GeForce RTX 3060]
• Any other relevant information: [e.g. I'm using a custom dataset]

Expected behavior

Model would train?

Screenshots

No response

Pip/GitHub

pip

What version/branch did you use?

1.0

Configuration YAML

/

Logs

Traceback (most recent call last):
  File "/home/citriot/steve/anomalib/train_test.py", line 25, in <module>
    i, train_data = next(enumerate(datamodule.train_dataloader()))
  File "/home/citriot/miniconda3/envs/anomalibv2/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 630, in __next__
    data = self._next_data()
  File "/home/citriot/miniconda3/envs/anomalibv2/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 1345, in _next_data
    return self._process_data(data)
  File "/home/citriot/miniconda3/envs/anomalibv2/lib/python3.10/site-packages/torch/utils/data/dataloader.py", line 1371, in _process_data
    data.reraise()
  File "/home/citriot/miniconda3/envs/anomalibv2/lib/python3.10/site-packages/torch/_utils.py", line 694, in reraise
    raise exception
RuntimeError: Caught RuntimeError in DataLoader worker process 0.
Original Traceback (most recent call last):
  File "/home/citriot/miniconda3/envs/anomalibv2/lib/python3.10/site-packages/torch/utils/data/_utils/worker.py", line 308, in _worker_loop
    data = fetcher.fetch(index)
  File "/home/citriot/miniconda3/envs/anomalibv2/lib/python3.10/site-packages/torch/utils/data/_utils/fetch.py", line 54, in fetch
    return self.collate_fn(data)
  File "/home/citriot/miniconda3/envs/anomalibv2/lib/python3.10/site-packages/torch/utils/data/_utils/collate.py", line 265, in default_collate
    return collate(batch, collate_fn_map=default_collate_fn_map)
  File "/home/citriot/miniconda3/envs/anomalibv2/lib/python3.10/site-packages/torch/utils/data/_utils/collate.py", line 127, in collate
    return elem_type({key: collate([d[key] for d in batch], collate_fn_map=collate_fn_map) for key in elem})
  File "/home/citriot/miniconda3/envs/anomalibv2/lib/python3.10/site-packages/torch/utils/data/_utils/collate.py", line 127, in <dictcomp>
    return elem_type({key: collate([d[key] for d in batch], collate_fn_map=collate_fn_map) for key in elem})
  File "/home/citriot/miniconda3/envs/anomalibv2/lib/python3.10/site-packages/torch/utils/data/_utils/collate.py", line 123, in collate
    return collate_fn_map[collate_type](batch, collate_fn_map=collate_fn_map)
  File "/home/citriot/miniconda3/envs/anomalibv2/lib/python3.10/site-packages/torch/utils/data/_utils/collate.py", line 162, in collate_tensor_fn
    return torch.stack(batch, 0, out=out)
  File "/home/citriot/miniconda3/envs/anomalibv2/lib/python3.10/site-packages/torchvision/tv_tensors/_tv_tensor.py", line 77, in __torch_function__
    output = func(*args, **kwargs or dict())
RuntimeError: stack expects each tensor to be equal size, but got [3, 512, 512] at entry 0 and [3, 522, 512] at entry 14

Code of Conduct

• I agree to follow this project's Code of Conduct

[JoannaCCJH]

Hi! I had similar issue before and solved it by adding image_size=(256,256) argument to Folder()

datamodule = Folder(
  name="hazelnut_toy",
  root="/home/citriot/Downloads/hazelnut_toy",
  normal_dir="good",
  abnormal_dir="crack",
  mask_dir="mask/crack",
  normal_split_ratio=0.2,
  image_size=(256,256)
)

[stevejonathancitriot]

Sure that works, but with a custom dataset with the image_size parameter added, I'm still getting similar errors. I did try using torchvision transform to resize the images but that didn't work

[praycxy]

Have you solved it? I'm also having this issue and hope I can get your help

[samet-akcay]

Sure that works, but with a custom dataset with the image_size parameter added, I'm still getting similar errors. I did try using torchvision transform to resize the images but that didn't work

@stevejonathancitriot, can you elaborate what sort of similar issues are you getting? I'm trying the reproduce the issue, but it works here

[samet-akcay]

When I try the setup you provided above, I can train and test the model

[samet-akcay]

This is because data modules expect transforms or image-size arguments.
https://github.com/openvinotoolkit/anomalib/blob/6840780bcbb83d6197e3899a6161ff827f033314/src/anomalib/data/image/folder.py#L399-L402

When none of them is passed, the transforms are read from the model.
https://github.com/openvinotoolkit/anomalib/blob/6840780bcbb83d6197e3899a6161ff827f033314/src/anomalib/models/image/padim/lightning_model.py#L128-L136

If you instantiate the datamodule without these arguments and play with the datamodule to check the dataset items, you would get the error you shared. However, if you create a model and engine and train, it should work since the transforms will be obtained from model.

Converting this to a Q&A since this is not an issue