Why EfficienAD become less effective after update the version?

[lzes]

This is the result now

here is the result used the version about one month ago

I only find that the batchsize changed from 1 to 32,is this the reason?

[abc-125]

Hello, yes, this is most probably the reason. The paper uses batch size 1.

[blaz-r]

I also checked the code and this is now part of EfficientAD:

anomalib/src/anomalib/models/image/efficient_ad/lightning_model.py

Lines 311 to 319 in 9482d34

	def configure_transforms(self, image_size: tuple[int, int] | None = None) -> Transform:
	"""Default transform for Padim."""
	image_size = image_size or (256, 256)
	return Compose(
	[
	Resize(image_size, antialias=True),
	Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
	],
	)

Isn't efficientAD supposed to work without imagenet normalization?

[abc-125]

Isn't efficientAD supposed to work without imagenet normalization?

Yes, there is an internal normalization. This transformation should have only Resize.