Batch Size Inference Warning during UniNet Training.

[1999swallowthemoon]

When training UniNet with MVTecAD (category: bottle), I'm encountering a warning about batch size inference even though train_batch_size=8 and eval_batch_size=8 are explicitly configured in the data module:
warning:lightning/pytorch/utilities/data.py:79: Trying to infer the batch_size from an ambiguous collection. The batch size we found is 3. To avoid any miscalculations, use self.log(..., batch_size=batch_size).

Reproduction Code:

from anomalib.data import MVTecAD
from anomalib.models import UniNet
from anomalib.engine import Engine

datamodule = MVTecAD(
    root="/datasets/MVTecAD',
    category="bottle",
    train_batch_size=8,
    eval_batch_size=8,
    seed=42,
)

pre_processor = UniNet.configure_pre_processor(image_size=(384, 384))
model = UniNet(pre_processor=pre_processor, post_processor=False)

engine = Engine(max_epochs=50)
engine.fit(datamodule=datamodule, model=model)

Do I need to follow the prompts to fix this warning?
Thank you for your guidance on this!

[waschsalz]

Hello @1999swallowthemoon

This warning is quite frequent but first and foremost it is neither a big problem nor something that you should fix.

It originates from the model implementation as the Lighting Trainer does not receive a clear value for the batch size, so it 'guesses'. I've only seen it guess correctly however this warning could be part of a pull request in the future.

[1999swallowthemoon]

Thanks for your answer!