🔨 Remove input_size argument from models

[samet-akcay]

Description

After merging PR #1706, models no longer require input_size argument. However, there are some places where input_size is still used, such as docs and some other utility functions. This is to ensure they are all removed properly.

For example, here is an example to the Patchcore model.

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

Lines 25 to 33 in 34b3a90

	class Patchcore(MemoryBankMixin, AnomalyModule):
	"""PatchcoreLightning Module to train PatchCore algorithm.
	Args:
	input_size (tuple[int, int]): Size of the model input.
	Defaults to ``(224, 224)``.
	backbone (str): Backbone CNN network
	Defaults to ``wide_resnet50_2``.
	layers (list[str]): Layers to extract features from the backbone CNN

[Shakib-IO]

Hi @samet-akcay
I want to work on this issue.

[samet-akcay]

Sure, thanks for your interest

[Shakib-IO]

Hi @samet-akcay,
I've begun addressing the issue at hand and have a few inquiries.

Certain utility functions contain the parameter input_size, as seen here (cfa). Should I also eliminate these instances?

Additionally, some functions compute values dependent on input_size, like here here (ganomaly) Should these calculations be adjusted as well?

Thanks

[dyogaharshitha]

.take

[Shakib-IO]

Hi @samet-akcay
In ganomaly, the encoder and decoder have the input_size, based on the input_size it calculates some values. Should these calculations be removed and adjusted as well? Please let me know.

Thanks.

[Shakib-IO]

Hi @blaz-r
Any comment on this?

[blaz-r]

Hi,
about the cfa question you had I think that this input_size stays as it's internal.

About ganomaly, the input_size you linked is part of torch model and that needs to stay. You should remove the input_size in lighting_model docstring:

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

Lines 26 to 31 in 34b3a90

	class Ganomaly(AnomalyModule):
	"""PL Lightning Module for the GANomaly Algorithm.
	Args:
	input_size (tuple[int, int]): Input dimension.
	Defaults to ``(256, 256)``.

the input_size is set inside lighting_model with setup function:

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

Lines 95 to 103 in 34b3a90

	def _setup(self) -> None:
	assert self.input_size is not None, "CSflow needs input size to build torch model."
	self.model = GanomalyModel(
	input_size=self.input_size,
	num_input_channels=3,
	n_features=self.n_features,
	latent_vec_size=self.latent_vec_size,
	extra_layers=self.extra_layers,
	add_final_conv_layer=self.add_final_conv_layer,

so the code you linked will work. But looking at this setup function, it seems like the assert string is wrong as it says csflow, so you can fix that.

[Shakib-IO]

Hi @blaz-r
Thanks.

Please review the PR and let me know.
Thank you.