From 8b4fca4da8a1b1bff2fbc2fe9131d019cdaf588e Mon Sep 17 00:00:00 2001
From: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
Date: Wed, 2 Jul 2025 17:00:31 +0100
Subject: [PATCH 01/16] inital commit: Don't re-init buffer, inplace fill

Signed-off-by: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
---
 src/anomalib/models/components/base/memory_bank_module.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/anomalib/models/components/base/memory_bank_module.py b/src/anomalib/models/components/base/memory_bank_module.py
index 07eae880bc..bf3a0d6d82 100644
--- a/src/anomalib/models/components/base/memory_bank_module.py
+++ b/src/anomalib/models/components/base/memory_bank_module.py
@@ -73,7 +73,7 @@ def on_validation_start(self) -> None:
         """
         if not self._is_fitted:
             self.fit()
-            self._is_fitted = torch.tensor([True], device=self.device)
+            self._is_fitted.data.fill_(value=True)
 
     def on_train_epoch_end(self) -> None:
         """Ensure memory bank is fitted after training.
@@ -82,4 +82,4 @@ def on_train_epoch_end(self) -> None:
         """
         if not self._is_fitted:
             self.fit()
-            self._is_fitted = torch.tensor([True], device=self.device)
+            self._is_fitted.data.fill_(value=True)

From 0c603c375fd33b3c2f1ba1408210fcc9a07866ef Mon Sep 17 00:00:00 2001
From: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
Date: Wed, 2 Jul 2025 20:22:26 +0100
Subject: [PATCH 02/16] implement patchcore memory savings

Signed-off-by: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
---
 .../models/image/patchcore/lightning_model.py | 13 ++------
 .../models/image/patchcore/torch_model.py     | 32 +++++++++++++------
 2 files changed, 25 insertions(+), 20 deletions(-)

diff --git a/src/anomalib/models/image/patchcore/lightning_model.py b/src/anomalib/models/image/patchcore/lightning_model.py
index ac673b85b2..8180d8497f 100644
--- a/src/anomalib/models/image/patchcore/lightning_model.py
+++ b/src/anomalib/models/image/patchcore/lightning_model.py
@@ -159,7 +159,6 @@ def __init__(
             num_neighbors=num_neighbors,
         )
         self.coreset_sampling_ratio = coreset_sampling_ratio
-        self.embeddings: list[torch.Tensor] = []
 
     @classmethod
     def configure_pre_processor(
@@ -235,9 +234,7 @@ def training_step(self, batch: Batch, *args, **kwargs) -> None:
             ``fit()``.
         """
         del args, kwargs  # These variables are not used.
-
-        embedding = self.model(batch.image)
-        self.embeddings.append(embedding)
+        _ = self.model(batch.image)
         # Return a dummy loss tensor
         return torch.tensor(0.0, requires_grad=True, device=self.device)
 
@@ -245,14 +242,10 @@ def fit(self) -> None:
         """Apply subsampling to the embedding collected from the training set.
 
         This method:
-        1. Aggregates embeddings from all training batches
-        2. Applies coreset subsampling to reduce memory requirements
+        1. Applies coreset subsampling to reduce memory requirements
         """
-        logger.info("Aggregating the embedding extracted from the training set.")
-        embeddings = torch.vstack(self.embeddings)
-
         logger.info("Applying core-set subsampling to get the embedding.")
-        self.model.subsample_embedding(embeddings, self.coreset_sampling_ratio)
+        self.model.subsample_embedding(self.coreset_sampling_ratio)
 
     def validation_step(self, batch: Batch, *args, **kwargs) -> STEP_OUTPUT:
         """Generate predictions for a batch of images.
diff --git a/src/anomalib/models/image/patchcore/torch_model.py b/src/anomalib/models/image/patchcore/torch_model.py
index 2a777a81ab..2ee7c86779 100644
--- a/src/anomalib/models/image/patchcore/torch_model.py
+++ b/src/anomalib/models/image/patchcore/torch_model.py
@@ -123,9 +123,7 @@ def __init__(
         ).eval()
         self.feature_pooler = torch.nn.AvgPool2d(3, 1, 1)
         self.anomaly_map_generator = AnomalyMapGenerator()
-
-        self.register_buffer("memory_bank", torch.Tensor())
-        self.memory_bank: torch.Tensor
+        self.register_buffer("memory_bank", torch.empty(0, self.feature_extractor.out_dims[1]))
 
     def forward(self, input_tensor: torch.Tensor) -> torch.Tensor | InferenceBatch:
         """Process input tensor through the model.
@@ -169,7 +167,19 @@ def forward(self, input_tensor: torch.Tensor) -> torch.Tensor | InferenceBatch:
         embedding = self.reshape_embedding(embedding)
 
         if self.training:
+            # Grow memory bank while preserving buffer registration
+            if self.memory_bank.numel() > 0:
+                new_bank = torch.cat((self.memory_bank, embedding), dim=0)
+                self.memory_bank.data = new_bank
+            else:
+                self.memory_bank.data = embedding
             return embedding
+
+        # Ensure memory bank is not empty
+        if self.memory_bank.size(0) == 0:
+            msg = "Memory bank is empty. Cannot perform coreset selection."
+            raise ValueError(msg)
+
         # apply nearest neighbor search
         patch_scores, locations = self.nearest_neighbors(embedding=embedding, n_neighbors=1)
         # reshape to batch dimension
@@ -239,26 +249,28 @@ def reshape_embedding(embedding: torch.Tensor) -> torch.Tensor:
         embedding_size = embedding.size(1)
         return embedding.permute(0, 2, 3, 1).reshape(-1, embedding_size)
 
-    def subsample_embedding(self, embedding: torch.Tensor, sampling_ratio: float) -> None:
-        """Subsample embeddings using coreset selection.
+    def subsample_embedding(self, sampling_ratio: float) -> None:
+        """Subsample the memory_banks embeddings using coreset selection.
 
         Uses k-center-greedy coreset subsampling to select a representative
         subset of patch embeddings to store in the memory bank.
 
         Args:
-            embedding (torch.Tensor): Embedding tensor to subsample from.
             sampling_ratio (float): Fraction of embeddings to keep, in range (0,1].
 
         Example:
-            >>> embedding = torch.randn(1000, 512)
-            >>> model.subsample_embedding(embedding, sampling_ratio=0.1)
+            >>> model.memory_bank = torch.randn(1000, 512)
+            >>> model.subsample_embedding(sampling_ratio=0.1)
             >>> model.memory_bank.shape
             torch.Size([100, 512])
         """
+        if self.memory_bank.size(0) == 0:
+            msg = "Memory bank is empty. Cannot perform coreset selection."
+            raise ValueError(msg)
         # Coreset Subsampling
-        sampler = KCenterGreedy(embedding=embedding, sampling_ratio=sampling_ratio)
+        sampler = KCenterGreedy(embedding=self.memory_bank, sampling_ratio=sampling_ratio)
         coreset = sampler.sample_coreset()
-        self.memory_bank = coreset
+        self.memory_bank.data = coreset.clone()
 
     @staticmethod
     def euclidean_dist(x: torch.Tensor, y: torch.Tensor) -> torch.Tensor:

From 73fd2eebcc3d2a1a19b751a894186c346837b5bf Mon Sep 17 00:00:00 2001
From: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
Date: Thu, 3 Jul 2025 13:26:19 +0100
Subject: [PATCH 03/16] deprecation warning for internal subsample_embedding of
 patchcore

Signed-off-by: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
---
 src/anomalib/models/image/patchcore/torch_model.py | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)

diff --git a/src/anomalib/models/image/patchcore/torch_model.py b/src/anomalib/models/image/patchcore/torch_model.py
index 2ee7c86779..12ff9d9a3d 100644
--- a/src/anomalib/models/image/patchcore/torch_model.py
+++ b/src/anomalib/models/image/patchcore/torch_model.py
@@ -33,6 +33,7 @@
 # Copyright (C) 2022-2025 Intel Corporation
 # SPDX-License-Identifier: Apache-2.0
 
+import warnings
 from collections.abc import Sequence
 from typing import TYPE_CHECKING
 
@@ -249,7 +250,7 @@ def reshape_embedding(embedding: torch.Tensor) -> torch.Tensor:
         embedding_size = embedding.size(1)
         return embedding.permute(0, 2, 3, 1).reshape(-1, embedding_size)
 
-    def subsample_embedding(self, sampling_ratio: float) -> None:
+    def subsample_embedding(self, sampling_ratio: float, embeddings: torch.Tensor = None) -> None:
         """Subsample the memory_banks embeddings using coreset selection.
 
         Uses k-center-greedy coreset subsampling to select a representative
@@ -257,6 +258,9 @@ def subsample_embedding(self, sampling_ratio: float) -> None:
 
         Args:
             sampling_ratio (float): Fraction of embeddings to keep, in range (0,1].
+            embeddings (torch.Tensor): **[DEPRECATED]**
+            This argument is deprecated and will be removed in a future release.
+            Use the default behavior (i.e., rely on `self.memory_bank`) instead.
 
         Example:
             >>> model.memory_bank = torch.randn(1000, 512)
@@ -264,6 +268,13 @@ def subsample_embedding(self, sampling_ratio: float) -> None:
             >>> model.memory_bank.shape
             torch.Size([100, 512])
         """
+        if embeddings is not None:
+            warnings.warn(
+                "The 'embeddings' argument is deprecated and will be removed in a future release."
+                "Please use the default memory bank instead.",
+                DeprecationWarning,
+                stacklevel=2,
+            )
         if self.memory_bank.size(0) == 0:
             msg = "Memory bank is empty. Cannot perform coreset selection."
             raise ValueError(msg)

From f2dca3141846359810f0c05359e89786d72a8885 Mon Sep 17 00:00:00 2001
From: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
Date: Thu, 3 Jul 2025 14:33:55 +0100
Subject: [PATCH 04/16] PaDiM gpu memory usage reduction

Signed-off-by: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
---
 .../models/image/padim/lightning_model.py     | 11 ++-----
 .../models/image/padim/torch_model.py         | 29 +++++++++++++++++++
 2 files changed, 31 insertions(+), 9 deletions(-)

diff --git a/src/anomalib/models/image/padim/lightning_model.py b/src/anomalib/models/image/padim/lightning_model.py
index c8f24b0b26..c890992f2e 100644
--- a/src/anomalib/models/image/padim/lightning_model.py
+++ b/src/anomalib/models/image/padim/lightning_model.py
@@ -131,9 +131,6 @@ def __init__(
             n_features=n_features,
         )
 
-        self.stats: list[torch.Tensor] = []
-        self.embeddings: list[torch.Tensor] = []
-
     @staticmethod
     def configure_optimizers() -> None:
         """PADIM doesn't require optimization, therefore returns no optimizers."""
@@ -154,19 +151,15 @@ def training_step(self, batch: Batch, *args, **kwargs) -> None:
         """
         del args, kwargs  # These variables are not used.
 
-        embedding = self.model(batch.image)
-        self.embeddings.append(embedding)
+        _ = self.model(batch.image)
 
         # Return a dummy loss tensor
         return torch.tensor(0.0, requires_grad=True, device=self.device)
 
     def fit(self) -> None:
         """Fit a Gaussian to the embedding collected from the training set."""
-        logger.info("Aggregating the embedding extracted from the training set.")
-        embeddings = torch.vstack(self.embeddings)
-
         logger.info("Fitting a Gaussian to the embedding collected from the training set.")
-        self.stats = self.model.gaussian.fit(embeddings)
+        self.model.fit_gaussian()
 
     def validation_step(self, batch: Batch, *args, **kwargs) -> STEP_OUTPUT:
         """Perform a validation step of PADIM.
diff --git a/src/anomalib/models/image/padim/torch_model.py b/src/anomalib/models/image/padim/torch_model.py
index 317299bdc6..3cc07bfefd 100644
--- a/src/anomalib/models/image/padim/torch_model.py
+++ b/src/anomalib/models/image/padim/torch_model.py
@@ -147,6 +147,7 @@ def __init__(
         self.anomaly_map_generator = AnomalyMapGenerator()
 
         self.gaussian = MultiVariateGaussian()
+        self.memory_bank = torch.empty(0, self.feature_extractor.out_dims[1])
 
     def forward(self, input_tensor: torch.Tensor) -> torch.Tensor | InferenceBatch:
         """Forward-pass image-batch (N, C, H, W) into model to extract features.
@@ -182,6 +183,12 @@ def forward(self, input_tensor: torch.Tensor) -> torch.Tensor | InferenceBatch:
             embeddings = self.tiler.untile(embeddings)
 
         if self.training:
+            # Grow memory bank
+            if self.memory_bank.numel() > 0:
+                new_bank = torch.cat((self.memory_bank, embeddings), dim=0)
+                self.memory_bank.data = new_bank
+            else:
+                self.memory_bank.data = embeddings
             return embeddings
 
         anomaly_map = self.anomaly_map_generator(
@@ -217,3 +224,25 @@ def generate_embedding(self, features: dict[str, torch.Tensor]) -> torch.Tensor:
         # subsample embeddings
         idx = self.idx.to(embeddings.device)
         return torch.index_select(embeddings, 1, idx)
+
+    def fit_gaussian(self) -> None:
+        """Fits a Gaussian model to the current contents of the memory bank.
+
+        This method is typically called after the memory bank has been filled during training.
+        It uses the stored features to fit a Gaussian distribution, which may be used for
+        downstream tasks such as anomaly detection, coreset selection, or probabilistic modeling.
+
+        After fitting, the memory bank is cleared to free GPU memory before validation or testing.
+
+        Raises:
+            ValueError: If the memory bank is empty.
+        """
+        if self.memory_bank.size(0) == 0:
+            msg = "Memory bank is empty. Cannot perform coreset selection."
+            raise ValueError(msg)
+
+        # fit gaussian
+        self.gaussian.fit(self.memory_bank)
+
+        # clear memory bank, redcues gpu size
+        self.memory_bank = torch.empty(0, self.feature_extractor.out_dims[1])

From 0275ce9616c3196186cc1b70e62f6c6bdf8299f1 Mon Sep 17 00:00:00 2001
From: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
Date: Thu, 3 Jul 2025 15:22:29 +0100
Subject: [PATCH 05/16] refactor memory bank tensor assignment for readability.
 dfkdde refactor

Signed-off-by: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
---
 .../models/image/dfkde/lightning_model.py     |  9 ++----
 .../models/image/dfkde/torch_model.py         | 28 +++++++++++++++++++
 .../models/image/padim/torch_model.py         | 17 ++++++-----
 .../models/image/patchcore/torch_model.py     | 18 ++++++------
 4 files changed, 48 insertions(+), 24 deletions(-)

diff --git a/src/anomalib/models/image/dfkde/lightning_model.py b/src/anomalib/models/image/dfkde/lightning_model.py
index dc8591eb54..2f4bd1aeea 100644
--- a/src/anomalib/models/image/dfkde/lightning_model.py
+++ b/src/anomalib/models/image/dfkde/lightning_model.py
@@ -113,8 +113,6 @@ def __init__(
             max_training_points=max_training_points,
         )
 
-        self.embeddings: list[torch.Tensor] = []
-
     @staticmethod
     def configure_optimizers() -> None:  # pylint: disable=arguments-differ
         """DFKDE doesn't require optimization, therefore returns no optimizers."""
@@ -133,18 +131,15 @@ def training_step(self, batch: Batch, *args, **kwargs) -> None:
         """
         del args, kwargs  # These variables are not used.
 
-        embedding = self.model(batch.image)
-        self.embeddings.append(embedding)
+        _ = self.model(batch.image)
 
         # Return a dummy loss tensor
         return torch.tensor(0.0, requires_grad=True, device=self.device)
 
     def fit(self) -> None:
         """Fit KDE model to collected embeddings from the training set."""
-        embeddings = torch.vstack(self.embeddings)
-
         logger.info("Fitting a KDE model to the embedding collected from the training set.")
-        self.model.classifier.fit(embeddings)
+        self.model.fit_classifier()
 
     def validation_step(self, batch: Batch, *args, **kwargs) -> STEP_OUTPUT:
         """Perform validation by computing anomaly scores.
diff --git a/src/anomalib/models/image/dfkde/torch_model.py b/src/anomalib/models/image/dfkde/torch_model.py
index bb7f9c28fe..b637a26f0c 100644
--- a/src/anomalib/models/image/dfkde/torch_model.py
+++ b/src/anomalib/models/image/dfkde/torch_model.py
@@ -89,6 +89,8 @@ def __init__(
             feature_scaling_method=feature_scaling_method,
             max_training_points=max_training_points,
         )
+        self.embedding_dim = sum(self.feature_extractor.out_dims)
+        self.memory_bank = torch.empty(0, self.embedding_dim)
 
     def get_features(self, batch: torch.Tensor) -> torch.Tensor:
         """Extract features from the pre-trained backbone network.
@@ -141,8 +143,34 @@ def forward(self, batch: torch.Tensor) -> torch.Tensor | InferenceBatch:
         # 1. apply feature extraction
         features = self.get_features(batch)
         if self.training:
+            with torch.no_grad():
+                if self.memory_bank.numel() > 0:
+                    self.memory_bank = torch.cat((self.memory_bank, features), dim=0)
+                else:
+                    self.memory_bank = features
             return features
 
         # 2. apply density estimation
         scores = self.classifier(features)
         return InferenceBatch(pred_score=scores)
+
+    def fit_classifier(self) -> None:
+        """Fits the classifier using the current contents of the memory bank.
+
+        This method is typically called after the memory bank has been populated
+        during training.
+
+        After fitting, the memory bank is cleared to reduce GPU memory usage.
+
+        Raises:
+            ValueError: If the memory bank is empty.
+        """
+        if self.memory_bank.size(0) == 0:
+            msg = "Memory bank is empty. Cannot perform coreset selection."
+            raise ValueError(msg)
+
+        # fit gaussian
+        self.classifier.fit(self.memory_bank)
+
+        # clear memory bank, redcues gpu size
+        self.memory_bank = torch.empty(0, self.embedding_dim)
diff --git a/src/anomalib/models/image/padim/torch_model.py b/src/anomalib/models/image/padim/torch_model.py
index 3cc07bfefd..5bc443dea8 100644
--- a/src/anomalib/models/image/padim/torch_model.py
+++ b/src/anomalib/models/image/padim/torch_model.py
@@ -147,7 +147,8 @@ def __init__(
         self.anomaly_map_generator = AnomalyMapGenerator()
 
         self.gaussian = MultiVariateGaussian()
-        self.memory_bank = torch.empty(0, self.feature_extractor.out_dims[1])
+        self.embedding_dim = sum(self.feature_extractor.out_dims)
+        self.memory_bank = torch.empty(0, self.embedding_dim)
 
     def forward(self, input_tensor: torch.Tensor) -> torch.Tensor | InferenceBatch:
         """Forward-pass image-batch (N, C, H, W) into model to extract features.
@@ -184,11 +185,11 @@ def forward(self, input_tensor: torch.Tensor) -> torch.Tensor | InferenceBatch:
 
         if self.training:
             # Grow memory bank
-            if self.memory_bank.numel() > 0:
-                new_bank = torch.cat((self.memory_bank, embeddings), dim=0)
-                self.memory_bank.data = new_bank
-            else:
-                self.memory_bank.data = embeddings
+            with torch.no_grad():
+                if self.memory_bank.numel() > 0:
+                    self.memory_bank = torch.cat((self.memory_bank, embeddings), dim=0)
+                else:
+                    self.memory_bank = embeddings
             return embeddings
 
         anomaly_map = self.anomaly_map_generator(
@@ -229,8 +230,6 @@ def fit_gaussian(self) -> None:
         """Fits a Gaussian model to the current contents of the memory bank.
 
         This method is typically called after the memory bank has been filled during training.
-        It uses the stored features to fit a Gaussian distribution, which may be used for
-        downstream tasks such as anomaly detection, coreset selection, or probabilistic modeling.
 
         After fitting, the memory bank is cleared to free GPU memory before validation or testing.
 
@@ -245,4 +244,4 @@ def fit_gaussian(self) -> None:
         self.gaussian.fit(self.memory_bank)
 
         # clear memory bank, redcues gpu size
-        self.memory_bank = torch.empty(0, self.feature_extractor.out_dims[1])
+        self.memory_bank = torch.empty(0, self.embedding_dim)
diff --git a/src/anomalib/models/image/patchcore/torch_model.py b/src/anomalib/models/image/patchcore/torch_model.py
index 12ff9d9a3d..5a52086c14 100644
--- a/src/anomalib/models/image/patchcore/torch_model.py
+++ b/src/anomalib/models/image/patchcore/torch_model.py
@@ -124,7 +124,9 @@ def __init__(
         ).eval()
         self.feature_pooler = torch.nn.AvgPool2d(3, 1, 1)
         self.anomaly_map_generator = AnomalyMapGenerator()
-        self.register_buffer("memory_bank", torch.empty(0, self.feature_extractor.out_dims[1]))
+        embedding_dim = sum(self.feature_extractor.out_dims)
+        self.memory_bank: torch.Tensor
+        self.register_buffer("memory_bank", torch.empty(0, embedding_dim))
 
     def forward(self, input_tensor: torch.Tensor) -> torch.Tensor | InferenceBatch:
         """Process input tensor through the model.
@@ -168,12 +170,12 @@ def forward(self, input_tensor: torch.Tensor) -> torch.Tensor | InferenceBatch:
         embedding = self.reshape_embedding(embedding)
 
         if self.training:
-            # Grow memory bank while preserving buffer registration
-            if self.memory_bank.numel() > 0:
-                new_bank = torch.cat((self.memory_bank, embedding), dim=0)
-                self.memory_bank.data = new_bank
-            else:
-                self.memory_bank.data = embedding
+            with torch.no_grad():
+                if self.memory_bank.numel() > 0:
+                    new_bank = torch.cat((self.memory_bank, embedding), dim=0)
+                    self.memory_bank.resize_(new_bank.size()).copy_(new_bank)
+                else:
+                    self.memory_bank.resize_(embedding.size()).copy_(embedding)
             return embedding
 
         # Ensure memory bank is not empty
@@ -281,7 +283,7 @@ def subsample_embedding(self, sampling_ratio: float, embeddings: torch.Tensor =
         # Coreset Subsampling
         sampler = KCenterGreedy(embedding=self.memory_bank, sampling_ratio=sampling_ratio)
         coreset = sampler.sample_coreset()
-        self.memory_bank.data = coreset.clone()
+        self.memory_bank = coreset
 
     @staticmethod
     def euclidean_dist(x: torch.Tensor, y: torch.Tensor) -> torch.Tensor:

From a0bdbf2ffe96a7604f9a520b59879c04a2bd5281 Mon Sep 17 00:00:00 2001
From: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
Date: Thu, 3 Jul 2025 15:49:40 +0100
Subject: [PATCH 06/16] unify memory bank models trainer arguments. refactor
 dfm to fit new memory bank framework

Signed-off-by: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
---
 .../models/image/dfkde/lightning_model.py     |  8 ++-
 .../models/image/dfm/lightning_model.py       | 20 +++----
 src/anomalib/models/image/dfm/torch_model.py  | 59 ++++++++++++++-----
 .../models/image/padim/lightning_model.py     | 14 ++---
 .../models/image/patchcore/lightning_model.py |  3 +-
 5 files changed, 67 insertions(+), 37 deletions(-)

diff --git a/src/anomalib/models/image/dfkde/lightning_model.py b/src/anomalib/models/image/dfkde/lightning_model.py
index 2f4bd1aeea..0c03415c49 100644
--- a/src/anomalib/models/image/dfkde/lightning_model.py
+++ b/src/anomalib/models/image/dfkde/lightning_model.py
@@ -162,9 +162,13 @@ def trainer_arguments(self) -> dict[str, Any]:
         """Get DFKDE-specific trainer arguments.
 
         Returns:
-            dict[str, Any]: Dictionary of trainer arguments.
+           dict[str, Any]: Trainer arguments
+               - ``gradient_clip_val``: ``0`` (no gradient clipping needed)
+               - ``max_epochs``: ``1`` (single pass through training data)
+               - ``num_sanity_val_steps``: ``0`` (skip validation sanity checks)
+               - ``devices``: ``1`` (only single gpu supported)
         """
-        return {"gradient_clip_val": 0, "max_epochs": 1, "num_sanity_val_steps": 0}
+        return {"gradient_clip_val": 0, "max_epochs": 1, "num_sanity_val_steps": 0, "devices": 1}
 
     @property
     def learning_type(self) -> LearningType:
diff --git a/src/anomalib/models/image/dfm/lightning_model.py b/src/anomalib/models/image/dfm/lightning_model.py
index bd0038dd77..9e128e00db 100644
--- a/src/anomalib/models/image/dfm/lightning_model.py
+++ b/src/anomalib/models/image/dfm/lightning_model.py
@@ -112,7 +112,6 @@ def __init__(
             n_comps=pca_level,
             score_type=score_type,
         )
-        self.embeddings: list[torch.Tensor] = []
         self.score_type = score_type
 
     @staticmethod
@@ -137,8 +136,7 @@ def training_step(self, batch: Batch, *args, **kwargs) -> None:
         """
         del args, kwargs  # These variables are not used.
 
-        embedding = self.model.get_features(batch.image).squeeze()
-        self.embeddings.append(embedding)
+        _ = self.model(batch.image)
 
         # Return a dummy loss tensor
         return torch.tensor(0.0, requires_grad=True, device=self.device)
@@ -149,11 +147,8 @@ def fit(self) -> None:
         The method aggregates embeddings collected during training and fits
         both the PCA transformation and Gaussian model used for scoring.
         """
-        logger.info("Aggregating the embedding extracted from the training set.")
-        embeddings = torch.vstack(self.embeddings)
-
         logger.info("Fitting a PCA and a Gaussian model to dataset.")
-        self.model.fit(embeddings)
+        self.model.fit()
 
     def validation_step(self, batch: Batch, *args, **kwargs) -> STEP_OUTPUT:
         """Compute predictions for the input batch during validation.
@@ -176,12 +171,13 @@ def trainer_arguments(self) -> dict[str, Any]:
         """Get DFM-specific trainer arguments.
 
         Returns:
-            dict[str, Any]: Dictionary of trainer arguments:
-                - ``gradient_clip_val`` (int): Disable gradient clipping
-                - ``max_epochs`` (int): Train for one epoch only
-                - ``num_sanity_val_steps`` (int): Skip validation sanity checks
+            dict[str, Any]: Trainer arguments
+                - ``gradient_clip_val``: ``0`` (no gradient clipping needed)
+                - ``max_epochs``: ``1`` (single pass through training data)
+                - ``num_sanity_val_steps``: ``0`` (skip validation sanity checks)
+                - ``devices``: ``1`` (only single gpu supported)
         """
-        return {"gradient_clip_val": 0, "max_epochs": 1, "num_sanity_val_steps": 0}
+        return {"gradient_clip_val": 0, "max_epochs": 1, "num_sanity_val_steps": 0, "devices": 1}
 
     @property
     def learning_type(self) -> LearningType:
diff --git a/src/anomalib/models/image/dfm/torch_model.py b/src/anomalib/models/image/dfm/torch_model.py
index 3e9aa8228a..664a28bae3 100644
--- a/src/anomalib/models/image/dfm/torch_model.py
+++ b/src/anomalib/models/image/dfm/torch_model.py
@@ -26,6 +26,7 @@
 # SPDX-License-Identifier: Apache-2.0
 
 import math
+import warnings
 
 import torch
 from torch import nn
@@ -153,18 +154,37 @@ def __init__(
             layers=[layer],
         ).eval()
 
-    def fit(self, dataset: torch.Tensor) -> None:
+        self.embedding_dim = sum(self.feature_extractor.out_dims)
+        self.memory_bank = torch.empty(0, self.embedding_dim)
+
+    def fit(self, dataset: torch.Tensor = None) -> None:
         """Fit PCA and Gaussian model to dataset.
 
         Args:
-            dataset (torch.Tensor): Input dataset with shape
-                ``(n_samples, n_features)``.
+            dataset (torch.Tensor, optional): **[DEPRECATED]**
+                Input dataset with shape ``(n_samples, n_features)``.
+                This argument is deprecated and will be ignored. Instead,
+                the method now uses `self.memory_bank` as the input dataset.
         """
-        self.pca_model.fit(dataset)
+        if dataset is not None:
+            warnings.warn(
+                "The 'dataset' argument is deprecated and will be removed in a future release. "
+                "The method now uses 'self.memory_bank' as the input dataset.",
+                DeprecationWarning,
+                stacklevel=2,
+            )
+
+        # Use self.memory_bank as the dataset
+        dataset_to_use = self.memory_bank
+
+        self.pca_model.fit(dataset_to_use)
         if self.score_type == "nll":
-            features_reduced = self.pca_model.transform(dataset)
+            features_reduced = self.pca_model.transform(dataset_to_use)
             self.gaussian_model.fit(features_reduced.T)
 
+        # clear memory bank, reduces GPU size
+        self.memory_bank = torch.empty(0, self.embedding_dim)
+
     def score(self, features: torch.Tensor, feature_shapes: tuple) -> torch.Tensor:
         """Compute anomaly scores.
 
@@ -202,17 +222,16 @@ def get_features(self, batch: torch.Tensor) -> torch.Tensor:
                 ``(batch_size, channels, height, width)``.
 
         Returns:
-            Union[torch.Tensor, Tuple[torch.Tensor, torch.Size]]: Features during
-                training, or tuple of (features, feature_shapes) during inference.
+            tuple of (features, feature_shapes).
         """
-        self.feature_extractor.eval()
-        features = self.feature_extractor(batch)[self.layer]
-        batch_size = len(features)
-        if self.pooling_kernel_size > 1:
-            features = F.avg_pool2d(input=features, kernel_size=self.pooling_kernel_size)
-        feature_shapes = features.shape
-        features = features.view(batch_size, -1).detach()
-        return features if self.training else (features, feature_shapes)
+        with torch.no_grad():
+            features = self.feature_extractor(batch)[self.layer]
+            batch_size = len(features)
+            if self.pooling_kernel_size > 1:
+                features = F.avg_pool2d(input=features, kernel_size=self.pooling_kernel_size)
+            feature_shapes = features.shape
+            features = features.view(batch_size, -1)
+        return features, feature_shapes
 
     def forward(self, batch: torch.Tensor) -> torch.Tensor | InferenceBatch:
         """Compute anomaly predictions from input images.
@@ -227,6 +246,16 @@ def forward(self, batch: torch.Tensor) -> torch.Tensor | InferenceBatch:
                 ``InferenceBatch`` with prediction scores and anomaly maps.
         """
         feature_vector, feature_shapes = self.get_features(batch)
+
+        if self.training:
+            # Grow memory bank
+            with torch.no_grad():
+                if self.memory_bank.numel() > 0:
+                    self.memory_bank = torch.cat((self.memory_bank, feature_vector), dim=0)
+                else:
+                    self.memory_bank = feature_vector
+            return feature_vector
+
         pred_score, anomaly_map = self.score(feature_vector.view(feature_vector.shape[:2]), feature_shapes)
         if anomaly_map is not None:
             anomaly_map = F.interpolate(anomaly_map, size=batch.shape[-2:], mode="bilinear", align_corners=False)
diff --git a/src/anomalib/models/image/padim/lightning_model.py b/src/anomalib/models/image/padim/lightning_model.py
index c890992f2e..57291dc8be 100644
--- a/src/anomalib/models/image/padim/lightning_model.py
+++ b/src/anomalib/models/image/padim/lightning_model.py
@@ -183,16 +183,16 @@ def validation_step(self, batch: Batch, *args, **kwargs) -> STEP_OUTPUT:
 
     @property
     def trainer_arguments(self) -> dict[str, int | float]:
-        """Return PADIM trainer arguments.
-
-        Since the model does not require training, we limit the max_epochs to 1.
-        Since we need to run training epoch before validation, we also set the
-        sanity steps to 0.
+        """Get default trainer arguments for Padim.
 
         Returns:
-            dict[str, int | float]: Dictionary of trainer arguments
+            dict[str, Any]: Trainer arguments
+                - ``max_epochs``: ``1`` (single pass through training data)
+                - ``val_check_interval``: ``1.0`` (check validation every 1 step)
+                - ``num_sanity_val_steps``: ``0`` (skip validation sanity checks)
+                - ``devices``: ``1`` (only single gpu supported)
         """
-        return {"max_epochs": 1, "val_check_interval": 1.0, "num_sanity_val_steps": 0}
+        return {"max_epochs": 1, "val_check_interval": 1.0, "num_sanity_val_steps": 0, "devices": 0}
 
     @property
     def learning_type(self) -> LearningType:
diff --git a/src/anomalib/models/image/patchcore/lightning_model.py b/src/anomalib/models/image/patchcore/lightning_model.py
index 8180d8497f..eec92feea5 100644
--- a/src/anomalib/models/image/patchcore/lightning_model.py
+++ b/src/anomalib/models/image/patchcore/lightning_model.py
@@ -279,8 +279,9 @@ def trainer_arguments(self) -> dict[str, Any]:
                 - ``gradient_clip_val``: ``0`` (no gradient clipping needed)
                 - ``max_epochs``: ``1`` (single pass through training data)
                 - ``num_sanity_val_steps``: ``0`` (skip validation sanity checks)
+                - ``devices``: ``1`` (only single gpu supported)
         """
-        return {"gradient_clip_val": 0, "max_epochs": 1, "num_sanity_val_steps": 0}
+        return {"gradient_clip_val": 0, "max_epochs": 1, "num_sanity_val_steps": 0, "devices": 1}
 
     @property
     def learning_type(self) -> LearningType:

From a3867328469936f8c7d9fbc8555651bb81a90422 Mon Sep 17 00:00:00 2001
From: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
Date: Thu, 3 Jul 2025 16:00:34 +0100
Subject: [PATCH 07/16] bugfix: padim device count set to zero

Signed-off-by: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
---
 src/anomalib/models/image/padim/lightning_model.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/anomalib/models/image/padim/lightning_model.py b/src/anomalib/models/image/padim/lightning_model.py
index 57291dc8be..0ad220b2c1 100644
--- a/src/anomalib/models/image/padim/lightning_model.py
+++ b/src/anomalib/models/image/padim/lightning_model.py
@@ -192,7 +192,7 @@ def trainer_arguments(self) -> dict[str, int | float]:
                 - ``num_sanity_val_steps``: ``0`` (skip validation sanity checks)
                 - ``devices``: ``1`` (only single gpu supported)
         """
-        return {"max_epochs": 1, "val_check_interval": 1.0, "num_sanity_val_steps": 0, "devices": 0}
+        return {"max_epochs": 1, "val_check_interval": 1.0, "num_sanity_val_steps": 0, "devices": 1}
 
     @property
     def learning_type(self) -> LearningType:

From 572dc5e19dc8de541b3820c66ce7b28976482d9c Mon Sep 17 00:00:00 2001
From: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
Date: Fri, 4 Jul 2025 15:34:15 +0100
Subject: [PATCH 08/16] ensure buffer is not replaced but instead resized and
 filled

Signed-off-by: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
---
 src/anomalib/models/image/patchcore/torch_model.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/anomalib/models/image/patchcore/torch_model.py b/src/anomalib/models/image/patchcore/torch_model.py
index 5a52086c14..3024a09e9a 100644
--- a/src/anomalib/models/image/patchcore/torch_model.py
+++ b/src/anomalib/models/image/patchcore/torch_model.py
@@ -283,7 +283,7 @@ def subsample_embedding(self, sampling_ratio: float, embeddings: torch.Tensor =
         # Coreset Subsampling
         sampler = KCenterGreedy(embedding=self.memory_bank, sampling_ratio=sampling_ratio)
         coreset = sampler.sample_coreset()
-        self.memory_bank = coreset
+        self.memory_bank.resize_(coreset.size()).copy_(coreset)
 
     @staticmethod
     def euclidean_dist(x: torch.Tensor, y: torch.Tensor) -> torch.Tensor:

From c714b85642cbdabb7b7c1b958134c2e60cc68379 Mon Sep 17 00:00:00 2001
From: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
Date: Mon, 7 Jul 2025 15:09:22 +0100
Subject: [PATCH 09/16] give memory bank type

Signed-off-by: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
---
 .../models/image/dfkde/torch_model.py         | 15 ++++++------
 src/anomalib/models/image/dfm/torch_model.py  | 23 ++++++++-----------
 .../models/image/padim/torch_model.py         | 18 +++++++--------
 .../models/image/patchcore/torch_model.py     | 20 ++++++++--------
 4 files changed, 33 insertions(+), 43 deletions(-)

diff --git a/src/anomalib/models/image/dfkde/torch_model.py b/src/anomalib/models/image/dfkde/torch_model.py
index b637a26f0c..8ff4aa1ec0 100644
--- a/src/anomalib/models/image/dfkde/torch_model.py
+++ b/src/anomalib/models/image/dfkde/torch_model.py
@@ -89,8 +89,7 @@ def __init__(
             feature_scaling_method=feature_scaling_method,
             max_training_points=max_training_points,
         )
-        self.embedding_dim = sum(self.feature_extractor.out_dims)
-        self.memory_bank = torch.empty(0, self.embedding_dim)
+        self.memory_bank = torch.Tensor()
 
     def get_features(self, batch: torch.Tensor) -> torch.Tensor:
         """Extract features from the pre-trained backbone network.
@@ -143,11 +142,11 @@ def forward(self, batch: torch.Tensor) -> torch.Tensor | InferenceBatch:
         # 1. apply feature extraction
         features = self.get_features(batch)
         if self.training:
-            with torch.no_grad():
-                if self.memory_bank.numel() > 0:
-                    self.memory_bank = torch.cat((self.memory_bank, features), dim=0)
-                else:
-                    self.memory_bank = features
+            if self.memory_bank.size(0) == 0:
+                self.memory_bank = features
+            else:
+                new_bank = torch.cat((self.memory_bank, features), dim=0).to(self.memory_bank)
+                self.memory_bank = new_bank
             return features
 
         # 2. apply density estimation
@@ -173,4 +172,4 @@ def fit_classifier(self) -> None:
         self.classifier.fit(self.memory_bank)
 
         # clear memory bank, redcues gpu size
-        self.memory_bank = torch.empty(0, self.embedding_dim)
+        self.memory_bank = torch.Tensor().to(self.memory_bank)
diff --git a/src/anomalib/models/image/dfm/torch_model.py b/src/anomalib/models/image/dfm/torch_model.py
index 664a28bae3..379256c5b7 100644
--- a/src/anomalib/models/image/dfm/torch_model.py
+++ b/src/anomalib/models/image/dfm/torch_model.py
@@ -154,8 +154,7 @@ def __init__(
             layers=[layer],
         ).eval()
 
-        self.embedding_dim = sum(self.feature_extractor.out_dims)
-        self.memory_bank = torch.empty(0, self.embedding_dim)
+        self.memory_bank = torch.Tensor()
 
     def fit(self, dataset: torch.Tensor = None) -> None:
         """Fit PCA and Gaussian model to dataset.
@@ -174,16 +173,13 @@ def fit(self, dataset: torch.Tensor = None) -> None:
                 stacklevel=2,
             )
 
-        # Use self.memory_bank as the dataset
-        dataset_to_use = self.memory_bank
-
-        self.pca_model.fit(dataset_to_use)
+        self.pca_model.fit(self.memory_bank)
         if self.score_type == "nll":
-            features_reduced = self.pca_model.transform(dataset_to_use)
+            features_reduced = self.pca_model.transform(self.memory_bank)
             self.gaussian_model.fit(features_reduced.T)
 
         # clear memory bank, reduces GPU size
-        self.memory_bank = torch.empty(0, self.embedding_dim)
+        self.memory_bank = torch.Tensor().to(self.memory_bank)
 
     def score(self, features: torch.Tensor, feature_shapes: tuple) -> torch.Tensor:
         """Compute anomaly scores.
@@ -248,12 +244,11 @@ def forward(self, batch: torch.Tensor) -> torch.Tensor | InferenceBatch:
         feature_vector, feature_shapes = self.get_features(batch)
 
         if self.training:
-            # Grow memory bank
-            with torch.no_grad():
-                if self.memory_bank.numel() > 0:
-                    self.memory_bank = torch.cat((self.memory_bank, feature_vector), dim=0)
-                else:
-                    self.memory_bank = feature_vector
+            if self.memory_bank.size(0) == 0:
+                self.memory_bank = feature_vector
+            else:
+                new_bank = torch.cat((self.memory_bank, feature_vector), dim=0).to(self.memory_bank)
+                self.memory_bank = new_bank
             return feature_vector
 
         pred_score, anomaly_map = self.score(feature_vector.view(feature_vector.shape[:2]), feature_shapes)
diff --git a/src/anomalib/models/image/padim/torch_model.py b/src/anomalib/models/image/padim/torch_model.py
index 5bc443dea8..6f88a57a09 100644
--- a/src/anomalib/models/image/padim/torch_model.py
+++ b/src/anomalib/models/image/padim/torch_model.py
@@ -147,8 +147,7 @@ def __init__(
         self.anomaly_map_generator = AnomalyMapGenerator()
 
         self.gaussian = MultiVariateGaussian()
-        self.embedding_dim = sum(self.feature_extractor.out_dims)
-        self.memory_bank = torch.empty(0, self.embedding_dim)
+        self.memory_bank = torch.Tensor()
 
     def forward(self, input_tensor: torch.Tensor) -> torch.Tensor | InferenceBatch:
         """Forward-pass image-batch (N, C, H, W) into model to extract features.
@@ -184,12 +183,11 @@ def forward(self, input_tensor: torch.Tensor) -> torch.Tensor | InferenceBatch:
             embeddings = self.tiler.untile(embeddings)
 
         if self.training:
-            # Grow memory bank
-            with torch.no_grad():
-                if self.memory_bank.numel() > 0:
-                    self.memory_bank = torch.cat((self.memory_bank, embeddings), dim=0)
-                else:
-                    self.memory_bank = embeddings
+            if self.memory_bank.size(0) == 0:
+                self.memory_bank = embeddings
+            else:
+                new_bank = torch.cat((self.memory_bank, embeddings), dim=0).to(self.memory_bank)
+                self.memory_bank = new_bank
             return embeddings
 
         anomaly_map = self.anomaly_map_generator(
@@ -243,5 +241,5 @@ def fit_gaussian(self) -> None:
         # fit gaussian
         self.gaussian.fit(self.memory_bank)
 
-        # clear memory bank, redcues gpu size
-        self.memory_bank = torch.empty(0, self.embedding_dim)
+        # clear memory bank, redcues gpu usage
+        self.memory_bank = torch.Tensor().to(self.memory_bank)
diff --git a/src/anomalib/models/image/patchcore/torch_model.py b/src/anomalib/models/image/patchcore/torch_model.py
index 3024a09e9a..8a65f01abc 100644
--- a/src/anomalib/models/image/patchcore/torch_model.py
+++ b/src/anomalib/models/image/patchcore/torch_model.py
@@ -124,9 +124,8 @@ def __init__(
         ).eval()
         self.feature_pooler = torch.nn.AvgPool2d(3, 1, 1)
         self.anomaly_map_generator = AnomalyMapGenerator()
-        embedding_dim = sum(self.feature_extractor.out_dims)
         self.memory_bank: torch.Tensor
-        self.register_buffer("memory_bank", torch.empty(0, embedding_dim))
+        self.register_buffer("memory_bank", torch.Tensor())
 
     def forward(self, input_tensor: torch.Tensor) -> torch.Tensor | InferenceBatch:
         """Process input tensor through the model.
@@ -170,17 +169,16 @@ def forward(self, input_tensor: torch.Tensor) -> torch.Tensor | InferenceBatch:
         embedding = self.reshape_embedding(embedding)
 
         if self.training:
-            with torch.no_grad():
-                if self.memory_bank.numel() > 0:
-                    new_bank = torch.cat((self.memory_bank, embedding), dim=0)
-                    self.memory_bank.resize_(new_bank.size()).copy_(new_bank)
-                else:
-                    self.memory_bank.resize_(embedding.size()).copy_(embedding)
+            if self.memory_bank.size(0) == 0:
+                self.memory_bank = embedding
+            else:
+                new_bank = torch.cat((self.memory_bank, embedding), dim=0).to(self.memory_bank)
+                self.memory_bank = new_bank
             return embedding
 
         # Ensure memory bank is not empty
         if self.memory_bank.size(0) == 0:
-            msg = "Memory bank is empty. Cannot perform coreset selection."
+            msg = "Memory bank is empty. Cannot provide anomaly scores"
             raise ValueError(msg)
 
         # apply nearest neighbor search
@@ -282,8 +280,8 @@ def subsample_embedding(self, sampling_ratio: float, embeddings: torch.Tensor =
             raise ValueError(msg)
         # Coreset Subsampling
         sampler = KCenterGreedy(embedding=self.memory_bank, sampling_ratio=sampling_ratio)
-        coreset = sampler.sample_coreset()
-        self.memory_bank.resize_(coreset.size()).copy_(coreset)
+        coreset = sampler.sample_coreset().type_as(self.memory_bank)
+        self.memory_bank = coreset
 
     @staticmethod
     def euclidean_dist(x: torch.Tensor, y: torch.Tensor) -> torch.Tensor:

From 16fabb31e8cde7e69be153dc8e718d058eaa86ef Mon Sep 17 00:00:00 2001
From: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
Date: Mon, 7 Jul 2025 15:13:09 +0100
Subject: [PATCH 10/16] revert memory bank mixin back

Signed-off-by: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
---
 src/anomalib/models/components/base/memory_bank_module.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/anomalib/models/components/base/memory_bank_module.py b/src/anomalib/models/components/base/memory_bank_module.py
index bf3a0d6d82..07eae880bc 100644
--- a/src/anomalib/models/components/base/memory_bank_module.py
+++ b/src/anomalib/models/components/base/memory_bank_module.py
@@ -73,7 +73,7 @@ def on_validation_start(self) -> None:
         """
         if not self._is_fitted:
             self.fit()
-            self._is_fitted.data.fill_(value=True)
+            self._is_fitted = torch.tensor([True], device=self.device)
 
     def on_train_epoch_end(self) -> None:
         """Ensure memory bank is fitted after training.
@@ -82,4 +82,4 @@ def on_train_epoch_end(self) -> None:
         """
         if not self._is_fitted:
             self.fit()
-            self._is_fitted.data.fill_(value=True)
+            self._is_fitted = torch.tensor([True], device=self.device)

From b292713160a286c8ed0414b35a294c88f192e4ce Mon Sep 17 00:00:00 2001
From: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
Date: Mon, 7 Jul 2025 15:54:47 +0100
Subject: [PATCH 11/16] to memory bank type and device

Signed-off-by: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
---
 src/anomalib/models/image/patchcore/torch_model.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/anomalib/models/image/patchcore/torch_model.py b/src/anomalib/models/image/patchcore/torch_model.py
index 8a65f01abc..0f5a1c742e 100644
--- a/src/anomalib/models/image/patchcore/torch_model.py
+++ b/src/anomalib/models/image/patchcore/torch_model.py
@@ -280,7 +280,7 @@ def subsample_embedding(self, sampling_ratio: float, embeddings: torch.Tensor =
             raise ValueError(msg)
         # Coreset Subsampling
         sampler = KCenterGreedy(embedding=self.memory_bank, sampling_ratio=sampling_ratio)
-        coreset = sampler.sample_coreset().type_as(self.memory_bank)
+        coreset = sampler.sample_coreset().to(self.memory_bank)
         self.memory_bank = coreset
 
     @staticmethod

From 585c81a6558773f0ee0bc08b509afc00eca62c75 Mon Sep 17 00:00:00 2001
From: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
Date: Fri, 11 Jul 2025 10:41:43 +0100
Subject: [PATCH 12/16] update return type, add comment about deprecation,
 duplicated copyright

Signed-off-by: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
---
 src/anomalib/models/image/dfm/torch_model.py       | 2 +-
 src/anomalib/models/image/patchcore/torch_model.py | 3 ---
 2 files changed, 1 insertion(+), 4 deletions(-)

diff --git a/src/anomalib/models/image/dfm/torch_model.py b/src/anomalib/models/image/dfm/torch_model.py
index de0930eaa5..a9e4986379 100644
--- a/src/anomalib/models/image/dfm/torch_model.py
+++ b/src/anomalib/models/image/dfm/torch_model.py
@@ -210,7 +210,7 @@ def score(self, features: torch.Tensor, feature_shapes: tuple) -> torch.Tensor:
 
         return (score, None) if self.score_type == "nll" else (score, score_map)
 
-    def get_features(self, batch: torch.Tensor) -> torch.Tensor:
+    def get_features(self, batch: torch.Tensor) -> tuple[torch.Tensor, torch.Size]:
         """Extract features from the pretrained network.
 
         Args:
diff --git a/src/anomalib/models/image/patchcore/torch_model.py b/src/anomalib/models/image/patchcore/torch_model.py
index 11d18aebf4..175e06baf0 100644
--- a/src/anomalib/models/image/patchcore/torch_model.py
+++ b/src/anomalib/models/image/patchcore/torch_model.py
@@ -33,9 +33,6 @@
         Coreset subsampling using k-center-greedy approach
 """
 
-# Copyright (C) 2022-2025 Intel Corporation
-# SPDX-License-Identifier: Apache-2.0
-
 import warnings
 from collections.abc import Sequence
 from typing import TYPE_CHECKING

From 9267dd19883192c07055043c4009402a3bc72a07 Mon Sep 17 00:00:00 2001
From: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
Date: Tue, 15 Jul 2025 11:21:18 +0100
Subject: [PATCH 13/16] new deprecation wrapper, now handles arg deprecation

Signed-off-by: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
---
 src/anomalib/utils/deprecation.py | 222 +++++++++++++++++++++++-------
 1 file changed, 175 insertions(+), 47 deletions(-)

diff --git a/src/anomalib/utils/deprecation.py b/src/anomalib/utils/deprecation.py
index 25717b8a50..e9986da6d4 100644
--- a/src/anomalib/utils/deprecation.py
+++ b/src/anomalib/utils/deprecation.py
@@ -3,7 +3,7 @@
 
 """Deprecation utilities for anomalib.
 
-This module provides utilities for marking functions and classes as deprecated.
+This module provides utilities for marking functions, classes and args as deprecated.
 The utilities help maintain backward compatibility while encouraging users to
 migrate to newer APIs.
 
@@ -34,6 +34,25 @@
     >>> @deprecate(since="2.1.0")
     ... def yet_another_function():
     ...     pass
+
+    >>> # Deprecation of specific argument(s)
+    >>> @deprecate(args={"old_param": "new_param"}, since="2.1.0", remove="3.0.0")
+    ... def my_function(new_param=None, old_param=None):
+    ...     # Handle the mapping logic yourself
+    ...     if old_param is not None and new_param is None:
+    ...         new_param = old_param
+    ...     # Rest of function logic
+    ...     pass
+
+    >>>   @deprecate(args={"old_param": "new_param", "two_param": "two_new_param"}, since="2.1.0")
+    >>>   def my_args_function(one_param, two_new_param=None, new_param=None, two_param=None, old_param=None):
+    ...     # Handle each mapping individually
+    ...     if old_param is not None and new_param is None:
+    ...         new_param = old_param
+    ...     if two_param is not None and two_new_param is None:
+    ...         two_new_param = two_param
+    ...     # Rest of function logic
+    ...     pass
 """
 
 import inspect
@@ -56,6 +75,7 @@ def deprecate(
     remove: str | None = None,
     use: str | None = None,
     reason: str | None = None,
+    args: dict[str, str] | None = None,
     warning_category: type[Warning] = FutureWarning,
 ) -> Callable[[_T], _T]: ...
 
@@ -67,25 +87,30 @@ def deprecate(
     remove: str | None = None,
     use: str | None = None,
     reason: str | None = None,
-    warning_category: type[Warning] = FutureWarning,
+    args: dict[str, str] | None = None,
+    warning_category: type[Warning] = DeprecationWarning,
 ) -> Any:
-    """Mark a function or class as deprecated.
+    """Mark a function, class, or keyword argument as deprecated.
 
-    This decorator will cause a warning to be emitted when the function is called or class is instantiated.
+    This decorator will cause a warning to be emitted when the function is called,
+    class is instantiated, or deprecated keyword arguments are used.
+    If args are passed, no function deprecation will be shown.
 
     Args:
         __obj: The function or class to be deprecated.
             If provided, the decorator is used without arguments.
             If None, the decorator is used with arguments.
-        since: Version when the function/class was deprecated (e.g. "2.1.0").
+        since: Version when the function/class/arg was deprecated (e.g. "2.1.0").
             If not provided, no deprecation version will be shown in the warning message.
-        remove: Version when the function/class will be removed.
+        remove: Version when the function/class/arg will be removed.
             If not provided, no removal version will be shown in the warning message.
-        use: Name of the replacement function/class.
+        use: Name of the replacement function/class/arg.
             If not provided, no replacement suggestion will be shown in the warning message.
         reason: Additional reason for the deprecation.
             If not provided, no reason will be shown in the warning message.
-        warning_category: Type of warning to emit (default: FutureWarning).
+        args: Mapping of deprecated argument names to their replacements.
+            If not provided, no argument deprecation warning will be shown.
+        warning_category: Type of warning to emit (default: DeprecationWarning).
             Can be DeprecationWarning, FutureWarning, PendingDeprecationWarning, etc.
 
     Returns:
@@ -116,49 +141,152 @@ def deprecate(
         >>> @deprecate(since="2.1.0", reason="Performance improvements available", warning_category=FutureWarning)
         ... def old_slow_function():
         ...     pass
+
+        >>> # Deprecation of specific argument(s)
+        >>> @deprecate(args={"old_param": "new_param"}, since="2.1.0", remove="3.0.0")
+        ... def my_function(new_param=None, old_param=None):
+        ...     # Handle the mapping logic yourself
+        ...     if old_param is not None and new_param is None:
+        ...         new_param = old_param
+        ...     # Rest of function logic
+        ...     pass
+
+        >>>   @deprecate(args={"old_param": "new_param", "two_param": "two_new_param"}, since="2.1.0")
+        >>>   def my_args_function(one_param, two_new_param=None, new_param=None, two_param=None, old_param=None):
+        ...     # Handle each mapping individually
+        ...     if old_param is not None and new_param is None:
+        ...         new_param = old_param
+        ...     if two_param is not None and two_new_param is None:
+        ...         two_new_param = two_param
+        ...     # Rest of function logic
+        ...     pass
     """
 
     def _deprecate_impl(obj: Any) -> Any:  # noqa: ANN401
         if isinstance(obj, type):
-            # Handle class deprecation
-            original_init = inspect.getattr_static(obj, "__init__")
-
-            @wraps(original_init)
-            def new_init(self: Any, *args: Any, **kwargs: Any) -> None:  # noqa: ANN401
-                msg = f"{obj.__name__} is deprecated"
-                if since:
-                    msg += f" since v{since}"
-                if remove:
-                    msg += f" and will be removed in v{remove}"
-                if use:
-                    msg += f". Use {use} instead"
-                if reason:
-                    msg += f". {reason}"
-                msg += "."
-                warnings.warn(msg, warning_category, stacklevel=2)
-                original_init(self, *args, **kwargs)
-
-            setattr(obj, "__init__", new_init)  # noqa: B010
-            return obj
-
-        # Handle function deprecation
-        @wraps(obj)
-        def wrapper(*args: Any, **kwargs: Any) -> Any:  # noqa: ANN401
-            msg = f"{obj.__name__} is deprecated"
-            if since:
-                msg += f" since v{since}"
-            if remove:
-                msg += f" and will be removed in v{remove}"
-            if use:
-                msg += f". Use {use} instead"
-            if reason:
-                msg += f". {reason}"
-            msg += "."
+            return _wrap_deprecated_init(obj, since, remove, use, reason, warning_category)
+        return _wrap_deprecated_function(obj, since, remove, use, reason, args, warning_category)
+
+    return _deprecate_impl if __obj is None else _deprecate_impl(__obj)
+
+
+def _build_deprecation_msg(
+    name: str,
+    since: str | None,
+    remove: str | None,
+    use: str | None,
+    reason: str | None,
+    prefix: str = "",
+) -> str:
+    """Construct a standardized deprecation warning message.
+
+    Args:
+        name: Name of the deprecated item (function, class, or argument).
+        since: Version when the item was deprecated.
+        remove: Version when the item will be removed.
+        use: Suggested replacement name.
+        reason: Additional reason for deprecation.
+        prefix: Optional prefix (e.g., 'Argument ') for clarity.
+
+    Returns:
+        A complete deprecation message string.
+    """
+    msg = f"{prefix}{name} is deprecated"
+    if since:
+        msg += f" since v{since}"
+    if remove:
+        msg += f" and will be removed in v{remove}"
+    if use:
+        msg += f". Use {use} instead"
+    if reason:
+        msg += f". {reason}"
+    return msg + "."
+
+
+def _wrap_deprecated_init(
+    cls: type,
+    since: str | None,
+    remove: str | None,
+    use: str | None,
+    reason: str | None,
+    warning_category: type[Warning],
+) -> type:
+    """Wrap a class constructor to emit a deprecation warning when instantiated.
+
+    Args:
+        cls: The class being deprecated.
+        since: Version when deprecation began.
+        remove: Version when the class will be removed.
+        use: Suggested replacement class.
+        reason: Additional reason for deprecation.
+        warning_category: The type of warning to emit.
+
+    Returns:
+        The decorated class with a warning-emitting __init__.
+    """
+    original_init = inspect.getattr_static(cls, "__init__")
+
+    @wraps(original_init)
+    def new_init(self: Any, *args: Any, **kwargs: Any) -> None:  # noqa: ANN401
+        msg = _build_deprecation_msg(cls.__name__, since, remove, use, reason)
+        warnings.warn(msg, warning_category, stacklevel=2)
+        original_init(self, *args, **kwargs)
+
+    setattr(cls, "__init__", new_init)  # noqa: B010
+    return cls
+
+
+def _wrap_deprecated_function(
+    func: Callable,
+    since: str | None,
+    remove: str | None,
+    use: str | None,
+    reason: str | None,
+    args_map: dict[str, str] | None,
+    warning_category: type[Warning],
+) -> Callable:
+    """Wrap a function to emit a deprecation warning when called.
+
+    Also handles deprecated keyword arguments, warning when old arguments are used.
+
+    Args:
+        func: The function to wrap.
+        since: Version when the function was deprecated.
+        remove: Version when the function will be removed.
+        use: Suggested replacement function.
+        reason: Additional reason for deprecation.
+        args_map: Mapping of deprecated argument names to new names.
+        warning_category: The type of warning to emit.
+
+    Returns:
+        The wrapped function that emits deprecation warnings when called.
+    """
+    sig = inspect.signature(func)
+
+    @wraps(func)
+    def wrapper(*args: Any, **kwargs: Any) -> Any:  # noqa: ANN401
+        bound_args = sig.bind_partial(*args, **kwargs)
+        bound_args.apply_defaults()
+
+        # Check for deprecated arguments and warn (but don't replace)
+        if args_map:
+            for old_arg, new_arg in args_map.items():
+                if old_arg in bound_args.arguments:
+                    msg = _build_deprecation_msg(
+                        old_arg,
+                        since,
+                        remove,
+                        use=new_arg,
+                        reason=None,
+                        prefix="Argument ",
+                    )
+                    warnings.warn(msg, warning_category, stacklevel=2)
+
+        # If args_map exists, this is argument deprecation, not function deprecation
+        if not args_map:
+            msg = _build_deprecation_msg(func.__name__, since, remove, use, reason)
             warnings.warn(msg, warning_category, stacklevel=2)
-            return obj(*args, **kwargs)
 
-        return wrapper
+        return func(*bound_args.args, **bound_args.kwargs)
 
-    if __obj is None:
-        return _deprecate_impl
-    return _deprecate_impl(__obj)
+    return wrapper

From d56a7bc44521b40cc8557b5f48e4c8e3f16d7c87 Mon Sep 17 00:00:00 2001
From: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
Date: Tue, 15 Jul 2025 11:54:02 +0100
Subject: [PATCH 14/16] add flexible tests for new deprecation warning

Signed-off-by: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
---
 src/anomalib/utils/deprecation.py  | 304 +++++++++++++++++------------
 tests/unit/utils/test_deprecate.py |  70 +++++++
 2 files changed, 245 insertions(+), 129 deletions(-)
 create mode 100644 tests/unit/utils/test_deprecate.py

diff --git a/src/anomalib/utils/deprecation.py b/src/anomalib/utils/deprecation.py
index e9986da6d4..baa91030ce 100644
--- a/src/anomalib/utils/deprecation.py
+++ b/src/anomalib/utils/deprecation.py
@@ -8,18 +8,11 @@
 migrate to newer APIs.
 
 Example:
-    >>> from anomalib.utils.deprecation import deprecated
-
-    >>> # Deprecation without any information
+    >>> # Basic deprecation without any information
     >>> @deprecate
     ... def old_function():
     ...     pass
 
-    >>> # Deprecation with replacement function
-    >>> @deprecate(use="new_function")
-    ... def old_function():
-    ...     pass
-
     >>> # Deprecation with version info and replacement class
     >>> @deprecate(since="2.1.0", remove="2.5.0", use="NewClass")
     ... class OldClass:
@@ -35,6 +28,17 @@
     ... def yet_another_function():
     ...     pass
 
+    >>> # Deprecation with reason and custom warning category
+    >>> @deprecate(since="2.1.0", reason="Performance improvements available", warning_category=FutureWarning)
+    ... def old_slow_function():
+    ...     pass
+
+    >>> # Deprecation of classes
+    >>> @deprecate(since="2.1.0", remove="3.0.0")
+    ... class MyClass:
+    ...     def __init__(self):
+    ...         pass
+
     >>> # Deprecation of specific argument(s)
     >>> @deprecate(args={"old_param": "new_param"}, since="2.1.0", remove="3.0.0")
     ... def my_function(new_param=None, old_param=None):
@@ -44,8 +48,8 @@
     ...     # Rest of function logic
     ...     pass
 
-    >>>   @deprecate(args={"old_param": "new_param", "two_param": "two_new_param"}, since="2.1.0")
-    >>>   def my_args_function(one_param, two_new_param=None, new_param=None, two_param=None, old_param=None):
+    >>> @deprecate(args={"old_param": "new_param", "two_param": "two_new_param"}, since="2.1.0")
+    >>> def my_args_function(one_param, two_new_param=None, new_param=None, two_param=None, old_param=None):
     ...     # Handle each mapping individually
     ...     if old_param is not None and new_param is None:
     ...         new_param = old_param
@@ -53,6 +57,13 @@
     ...         two_new_param = two_param
     ...     # Rest of function logic
     ...     pass
+
+    >>> @deprecate(args={"old_arg": "new_arg"}, since="2.1.0", remove="3.0.0")
+    >>> class MyClass:
+    ...     def __init__(self, new_arg=None, old_arg=None):
+    ...         if new_arg is None and old_arg is not None:
+    ...            new_arg = old_arg
+    ...         pass
 """
 
 import inspect
@@ -64,112 +75,6 @@
 _T = TypeVar("_T")
 
 
-@overload
-def deprecate(__obj: _T) -> _T: ...
-
-
-@overload
-def deprecate(
-    *,
-    since: str | None = None,
-    remove: str | None = None,
-    use: str | None = None,
-    reason: str | None = None,
-    args: dict[str, str] | None = None,
-    warning_category: type[Warning] = FutureWarning,
-) -> Callable[[_T], _T]: ...
-
-
-def deprecate(
-    __obj: Any = None,
-    *,
-    since: str | None = None,
-    remove: str | None = None,
-    use: str | None = None,
-    reason: str | None = None,
-    args: dict[str, str] | None = None,
-    warning_category: type[Warning] = DeprecationWarning,
-) -> Any:
-    """Mark a function, class, or keyword argument as deprecated.
-
-    This decorator will cause a warning to be emitted when the function is called,
-    class is instantiated, or deprecated keyword arguments are used.
-    If args are passed, no function deprecation will be shown.
-
-    Args:
-        __obj: The function or class to be deprecated.
-            If provided, the decorator is used without arguments.
-            If None, the decorator is used with arguments.
-        since: Version when the function/class/arg was deprecated (e.g. "2.1.0").
-            If not provided, no deprecation version will be shown in the warning message.
-        remove: Version when the function/class/arg will be removed.
-            If not provided, no removal version will be shown in the warning message.
-        use: Name of the replacement function/class/arg.
-            If not provided, no replacement suggestion will be shown in the warning message.
-        reason: Additional reason for the deprecation.
-            If not provided, no reason will be shown in the warning message.
-        args: Mapping of deprecated argument names to their replacements.
-            If not provided, no argument deprecation warning will be shown.
-        warning_category: Type of warning to emit (default: DeprecationWarning).
-            Can be DeprecationWarning, FutureWarning, PendingDeprecationWarning, etc.
-
-    Returns:
-        Decorated function/class that emits a deprecation warning when used.
-
-    Example:
-        >>> # Basic deprecation without any information
-        >>> @deprecate
-        ... def old_function():
-        ...     pass
-
-        >>> # Deprecation with version info and replacement class
-        >>> @deprecate(since="2.1.0", remove="2.5.0", use="NewClass")
-        ... class OldClass:
-        ...     pass
-
-        >>> # Deprecation with only removal version
-        >>> @deprecate(remove="2.5.0")
-        ... def another_function():
-        ...     pass
-
-        >>> # Deprecation with only deprecation version
-        >>> @deprecate(since="2.1.0")
-        ... def yet_another_function():
-        ...     pass
-
-        >>> # Deprecation with reason and custom warning category
-        >>> @deprecate(since="2.1.0", reason="Performance improvements available", warning_category=FutureWarning)
-        ... def old_slow_function():
-        ...     pass
-
-        >>> # Deprecation of specific argument(s)
-        >>> @deprecate(args={"old_param": "new_param"}, since="2.1.0", remove="3.0.0")
-        ... def my_function(new_param=None, old_param=None):
-        ...     # Handle the mapping logic yourself
-        ...     if old_param is not None and new_param is None:
-        ...         new_param = old_param
-        ...     # Rest of function logic
-        ...     pass
-
-        >>>   @deprecate(args={"old_param": "new_param", "two_param": "two_new_param"}, since="2.1.0")
-        >>>   def my_args_function(one_param, two_new_param=None, new_param=None, two_param=None, old_param=None):
-        ...     # Handle each mapping individually
-        ...     if old_param is not None and new_param is None:
-        ...         new_param = old_param
-        ...     if two_param is not None and two_new_param is None:
-        ...         two_new_param = two_param
-        ...     # Rest of function logic
-        ...     pass
-    """
-
-    def _deprecate_impl(obj: Any) -> Any:  # noqa: ANN401
-        if isinstance(obj, type):
-            return _wrap_deprecated_init(obj, since, remove, use, reason, warning_category)
-        return _wrap_deprecated_function(obj, since, remove, use, reason, args, warning_category)
-
-    return _deprecate_impl if __obj is None else _deprecate_impl(__obj)
-
-
 def _build_deprecation_msg(
     name: str,
     since: str | None,
@@ -203,12 +108,42 @@ def _build_deprecation_msg(
     return msg + "."
 
 
+def _warn_deprecated_arguments(
+    args_map: dict[str, str],
+    bound_args: inspect.BoundArguments,
+    since: str | None,
+    remove: str | None,
+    warning_category: type[Warning],
+) -> None:
+    """Warn about deprecated keyword arguments.
+
+    Args:
+        args_map: Mapping of deprecated argument names to new names.
+        bound_args: Bound arguments of the function or method.
+        since: Version when the argument was deprecated.
+        remove: Version when the argument will be removed.
+        warning_category: Type of warning to emit.
+    """
+    for old_arg, new_arg in args_map.items():
+        if old_arg in bound_args.arguments:
+            msg = _build_deprecation_msg(
+                old_arg,
+                since,
+                remove,
+                use=new_arg,
+                reason=None,
+                prefix="Argument ",
+            )
+            warnings.warn(msg, warning_category, stacklevel=3)
+
+
 def _wrap_deprecated_init(
     cls: type,
     since: str | None,
     remove: str | None,
     use: str | None,
     reason: str | None,
+    args_map: dict[str, str] | None,
     warning_category: type[Warning],
 ) -> type:
     """Wrap a class constructor to emit a deprecation warning when instantiated.
@@ -219,17 +154,26 @@ def _wrap_deprecated_init(
         remove: Version when the class will be removed.
         use: Suggested replacement class.
         reason: Additional reason for deprecation.
+        args_map: Mapping of deprecated argument names to new names.
         warning_category: The type of warning to emit.
 
     Returns:
         The decorated class with a warning-emitting __init__.
     """
     original_init = inspect.getattr_static(cls, "__init__")
+    sig = inspect.signature(original_init)
 
     @wraps(original_init)
     def new_init(self: Any, *args: Any, **kwargs: Any) -> None:  # noqa: ANN401
-        msg = _build_deprecation_msg(cls.__name__, since, remove, use, reason)
-        warnings.warn(msg, warning_category, stacklevel=2)
+        bound_args = sig.bind_partial(self, *args, **kwargs)
+        bound_args.apply_defaults()
+
+        if args_map:
+            _warn_deprecated_arguments(args_map, bound_args, since, remove, warning_category)
+        else:
+            msg = _build_deprecation_msg(cls.__name__, since, remove, use, reason)
+            warnings.warn(msg, warning_category, stacklevel=2)
+
         original_init(self, *args, **kwargs)
 
     setattr(cls, "__init__", new_init)  # noqa: B010
@@ -270,17 +214,7 @@ def wrapper(*args: Any, **kwargs: Any) -> Any:  # noqa: ANN401
 
         # Check for deprecated arguments and warn (but don't replace)
         if args_map:
-            for old_arg, new_arg in args_map.items():
-                if old_arg in bound_args.arguments:
-                    msg = _build_deprecation_msg(
-                        old_arg,
-                        since,
-                        remove,
-                        use=new_arg,
-                        reason=None,
-                        prefix="Argument ",
-                    )
-                    warnings.warn(msg, warning_category, stacklevel=2)
+            _warn_deprecated_arguments(args_map, bound_args, since, remove, warning_category)
 
         # If args_map exists, this is argument deprecation, not function deprecation
         if not args_map:
@@ -290,3 +224,115 @@ def wrapper(*args: Any, **kwargs: Any) -> Any:  # noqa: ANN401
         return func(*bound_args.args, **bound_args.kwargs)
 
     return wrapper
+
+
+@overload
+def deprecate(__obj: _T) -> _T: ...
+
+
+@overload
+def deprecate(
+    *,
+    since: str | None = None,
+    remove: str | None = None,
+    use: str | None = None,
+    reason: str | None = None,
+    args: dict[str, str] | None = None,
+    warning_category: type[Warning] = FutureWarning,
+) -> Callable[[_T], _T]: ...
+
+
+def deprecate(
+    __obj: Any = None,
+    *,
+    since: str | None = None,
+    remove: str | None = None,
+    use: str | None = None,
+    reason: str | None = None,
+    args: dict[str, str] | None = None,
+    warning_category: type[Warning] = DeprecationWarning,
+) -> Any:
+    """Mark a function, class, or keyword argument as deprecated.
+
+    This decorator will cause a warning to be emitted when the function is called,
+    class is instantiated, or deprecated keyword arguments are used.
+    If args are passed, no function deprecation will be shown.
+
+    Args:
+        __obj: The function or class to be deprecated.
+            If provided, the decorator is used without arguments.
+            If None, the decorator is used with arguments.
+        since: Version when the function/class/arg was deprecated (e.g. "2.1.0").
+            If not provided, no deprecation version will be shown in the warning message.
+        remove: Version when the function/class/arg will be removed.
+            If not provided, no removal version will be shown in the warning message.
+        use: Name of the replacement function/class/arg.
+            If not provided, no replacement suggestion will be shown in the warning message.
+        reason: Additional reason for the deprecation.
+            If not provided, no reason will be shown in the warning message.
+        args: Mapping of deprecated argument names to their replacements.
+            If not provided, no argument deprecation warning will be shown.
+            If provided, function deprecation will not be show.
+        warning_category: Type of warning to emit (default: DeprecationWarning).
+            Can be DeprecationWarning, FutureWarning, PendingDeprecationWarning, etc.
+
+    Returns:
+        Decorated function/class that emits a deprecation warning when used.
+
+    Example:
+        >>> # Basic deprecation without any information
+        >>> @deprecate
+        ... def old_function():
+        ...     pass
+
+        >>> # Deprecation with version info and replacement class
+        >>> @deprecate(since="2.1.0", remove="2.5.0", use="NewClass")
+        ... class OldClass:
+        ...     pass
+
+        >>> # Deprecation with only removal version
+        >>> @deprecate(remove="2.5.0")
+        ... def another_function():
+        ...     pass
+
+        >>> # Deprecation with only deprecation version
+        >>> @deprecate(since="2.1.0")
+        ... def yet_another_function():
+        ...     pass
+
+        >>> # Deprecation with reason and custom warning category
+        >>> @deprecate(since="2.1.0", reason="Performance improvements available", warning_category=FutureWarning)
+        ... def old_slow_function():
+        ...     pass
+
+        >>> # Deprecation of specific argument(s)
+        >>> @deprecate(args={"old_param": "new_param"}, since="2.1.0", remove="3.0.0")
+        ... def my_function(new_param=None, old_param=None):
+        ...     # Handle the mapping logic yourself
+        ...     if old_param is not None and new_param is None:
+        ...         new_param = old_param
+        ...     # Rest of function logic
+        ...     pass
+
+        >>> @deprecate(args={"old_param": "new_param", "two_param": "two_new_param"}, since="2.1.0")
+        >>> def my_args_function(one_param, two_new_param=None, new_param=None, two_param=None, old_param=None):
+        ...     # Handle each mapping individually
+        ...     if old_param is not None and new_param is None:
+        ...         new_param = old_param
+        ...     if two_param is not None and two_new_param is None:
+        ...         two_new_param = two_param
+        ...     # Rest of function logic
+        ...     pass
+        >>> @deprecate(args={"old_arg": "new_arg"}, since="2.1.0", remove="3.0.0")
+        >>> class MyClass:
+        >>>     def __init__(self, new_arg=None, old_arg=None):
+        >>>         if new_arg is None and old_arg is not None:
+        >>>            new_arg = old_arg
+    """
+
+    def _deprecate_impl(obj: Any) -> Any:  # noqa: ANN401
+        if isinstance(obj, type):
+            return _wrap_deprecated_init(obj, since, remove, use, reason, args, warning_category)
+        return _wrap_deprecated_function(obj, since, remove, use, reason, args, warning_category)
+
+    return _deprecate_impl if __obj is None else _deprecate_impl(__obj)
diff --git a/tests/unit/utils/test_deprecate.py b/tests/unit/utils/test_deprecate.py
new file mode 100644
index 0000000000..5b4e98589f
--- /dev/null
+++ b/tests/unit/utils/test_deprecate.py
@@ -0,0 +1,70 @@
+# Copyright (C) 2022-2025 Intel Corporation
+# SPDX-License-Identifier: Apache-2.0
+
+"""Tests for the deprecation utility function."""
+
+import pytest
+
+from anomalib.utils import deprecate
+
+
+def test_deprecated_function_warns() -> None:
+    """Test simple function deprecation warning."""
+
+    @deprecate(since="1.0", remove="2.0", use="new_func")
+    def old_func() -> str:
+        return "hello"
+
+    with pytest.warns(DeprecationWarning, match=r"(?=.*1\.0)(?=.*2\.0)(?=.*new_func)"):
+        assert old_func() == "hello"
+
+
+def test_deprecated_class_warns_on_init() -> None:
+    """Test deprecated class emits warning on instantiation."""
+
+    @deprecate(since="1.0", remove="2.0", use="NewClass")
+    class OldClass:
+        def __init__(self) -> None:
+            self.val = 42
+
+    with pytest.warns(DeprecationWarning, match=r"(?=.*1\.0)(?=.*2\.0)(?=.*NewClass)"):
+        instance = OldClass()
+    assert instance.val == 42
+
+
+def test_deprecated_function_arg_warns() -> None:
+    """Test deprecated argument triggers warning with correct match."""
+
+    @deprecate(args={"old_param": "new_param"}, since="1.0", remove="2.0")
+    def func(new_param: int | None = None, old_param: int | None = None) -> int:
+        return new_param if new_param is not None else old_param
+
+    with pytest.warns(DeprecationWarning, match=r"(?=.*old_param)(?=.*new_param)(?=.*1\.0)(?=.*2\.0)"):
+        assert func(old_param=10) == 10
+
+
+def test_deprecated_multiple_args_warns() -> None:
+    """Test multiple deprecated arguments trigger appropriate warnings."""
+
+    @deprecate(args={"old1": "new1", "old2": "new2"}, since="1.0")
+    def func(new1=None, new2=None, old1=None, old2=None):  # noqa: ANN001, ANN202
+        return new1 or old1, new2 or old2
+
+    with pytest.warns(DeprecationWarning, match=r"(?=.*old1)(?=.*new1)(?=.*1\.0)"):
+        func(old1="a")
+
+    with pytest.warns(DeprecationWarning, match=r"(?=.*old2)(?=.*new2)(?=.*1\.0)"):
+        func(old2="b")
+
+
+def test_deprecated_class_arg_warns() -> None:
+    """Test deprecated constructor argument triggers warning."""
+
+    @deprecate(args={"legacy_arg": "modern_arg"}, since="1.0", remove="2.0")
+    class MyClass:
+        def __init__(self, modern_arg=None, legacy_arg=None) -> None:  # noqa: ANN001
+            self.value = modern_arg if modern_arg is not None else legacy_arg
+
+    with pytest.warns(DeprecationWarning, match=r"(?=.*legacy_arg)(?=.*modern_arg)(?=.*1\.0)(?=.*2\.0)"):
+        obj = MyClass(legacy_arg="deprecated")
+    assert obj.value == "deprecated"

From 9072bbdee4bf3514989b2c0849f357215c5f596b Mon Sep 17 00:00:00 2001
From: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
Date: Tue, 15 Jul 2025 12:12:59 +0100
Subject: [PATCH 15/16] update warnings to deprecations. Resolve comments
 re-fit vs fit_guassian. Test for None replacements for args

Signed-off-by: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
---
 src/anomalib/models/image/dfkde/lightning_model.py |  2 +-
 src/anomalib/models/image/dfkde/torch_model.py     |  6 +++---
 src/anomalib/models/image/dfm/torch_model.py       | 14 +++++---------
 src/anomalib/models/image/padim/lightning_model.py |  2 +-
 src/anomalib/models/image/padim/torch_model.py     |  6 +++---
 src/anomalib/models/image/patchcore/torch_model.py | 13 +++++--------
 src/anomalib/utils/deprecation.py                  | 10 +++++-----
 tests/unit/utils/test_deprecate.py                 | 11 +++++++++++
 8 files changed, 34 insertions(+), 30 deletions(-)

diff --git a/src/anomalib/models/image/dfkde/lightning_model.py b/src/anomalib/models/image/dfkde/lightning_model.py
index 0b57e3c9ab..99ac990e1d 100644
--- a/src/anomalib/models/image/dfkde/lightning_model.py
+++ b/src/anomalib/models/image/dfkde/lightning_model.py
@@ -139,7 +139,7 @@ def training_step(self, batch: Batch, *args, **kwargs) -> None:
     def fit(self) -> None:
         """Fit KDE model to collected embeddings from the training set."""
         logger.info("Fitting a KDE model to the embedding collected from the training set.")
-        self.model.fit_classifier()
+        self.model.fit()
 
     def validation_step(self, batch: Batch, *args, **kwargs) -> STEP_OUTPUT:
         """Perform validation by computing anomaly scores.
diff --git a/src/anomalib/models/image/dfkde/torch_model.py b/src/anomalib/models/image/dfkde/torch_model.py
index c797889b4a..9af1ef4b65 100644
--- a/src/anomalib/models/image/dfkde/torch_model.py
+++ b/src/anomalib/models/image/dfkde/torch_model.py
@@ -89,7 +89,7 @@ def __init__(
             feature_scaling_method=feature_scaling_method,
             max_training_points=max_training_points,
         )
-        self.memory_bank = torch.Tensor()
+        self.memory_bank = torch.empty(0)
 
     def get_features(self, batch: torch.Tensor) -> torch.Tensor:
         """Extract features from the pre-trained backbone network.
@@ -153,7 +153,7 @@ def forward(self, batch: torch.Tensor) -> torch.Tensor | InferenceBatch:
         scores = self.classifier(features)
         return InferenceBatch(pred_score=scores)
 
-    def fit_classifier(self) -> None:
+    def fit(self) -> None:
         """Fits the classifier using the current contents of the memory bank.
 
         This method is typically called after the memory bank has been populated
@@ -172,4 +172,4 @@ def fit_classifier(self) -> None:
         self.classifier.fit(self.memory_bank)
 
         # clear memory bank, redcues gpu size
-        self.memory_bank = torch.Tensor().to(self.memory_bank)
+        self.memory_bank = torch.empty(0).to(self.memory_bank)
diff --git a/src/anomalib/models/image/dfm/torch_model.py b/src/anomalib/models/image/dfm/torch_model.py
index a9e4986379..25cbb2c36a 100644
--- a/src/anomalib/models/image/dfm/torch_model.py
+++ b/src/anomalib/models/image/dfm/torch_model.py
@@ -26,7 +26,6 @@
 """
 
 import math
-import warnings
 
 import torch
 from torch import nn
@@ -34,6 +33,7 @@
 
 from anomalib.data import InferenceBatch
 from anomalib.models.components import PCA, DynamicBufferMixin, TimmFeatureExtractor
+from anomalib.utils.deprecation import deprecate
 
 
 class SingleClassGaussian(DynamicBufferMixin):
@@ -154,8 +154,9 @@ def __init__(
             layers=[layer],
         ).eval()
 
-        self.memory_bank = torch.Tensor()
+        self.memory_bank = torch.empty(0)
 
+    @deprecate(args={"dataset": None}, since="2.1.0")
     def fit(self, dataset: torch.Tensor = None) -> None:
         """Fit PCA and Gaussian model to dataset.
 
@@ -166,12 +167,7 @@ def fit(self, dataset: torch.Tensor = None) -> None:
                 the method now uses `self.memory_bank` as the input dataset.
         """
         if dataset is not None:
-            warnings.warn(
-                "The 'dataset' argument is deprecated and will be removed in a future release. "
-                "The method now uses 'self.memory_bank' as the input dataset.",
-                DeprecationWarning,
-                stacklevel=2,
-            )
+            del dataset
 
         self.pca_model.fit(self.memory_bank)
         if self.score_type == "nll":
@@ -179,7 +175,7 @@ def fit(self, dataset: torch.Tensor = None) -> None:
             self.gaussian_model.fit(features_reduced.T)
 
         # clear memory bank, reduces GPU size
-        self.memory_bank = torch.Tensor().to(self.memory_bank)
+        self.memory_bank = torch.empty(0).to(self.memory_bank)
 
     def score(self, features: torch.Tensor, feature_shapes: tuple) -> torch.Tensor:
         """Compute anomaly scores.
diff --git a/src/anomalib/models/image/padim/lightning_model.py b/src/anomalib/models/image/padim/lightning_model.py
index 7e76873467..8660b88ee4 100644
--- a/src/anomalib/models/image/padim/lightning_model.py
+++ b/src/anomalib/models/image/padim/lightning_model.py
@@ -159,7 +159,7 @@ def training_step(self, batch: Batch, *args, **kwargs) -> None:
     def fit(self) -> None:
         """Fit a Gaussian to the embedding collected from the training set."""
         logger.info("Fitting a Gaussian to the embedding collected from the training set.")
-        self.model.fit_gaussian()
+        self.model.fit()
 
     def validation_step(self, batch: Batch, *args, **kwargs) -> STEP_OUTPUT:
         """Perform a validation step of PADIM.
diff --git a/src/anomalib/models/image/padim/torch_model.py b/src/anomalib/models/image/padim/torch_model.py
index 0d08a3990a..d7367149a0 100644
--- a/src/anomalib/models/image/padim/torch_model.py
+++ b/src/anomalib/models/image/padim/torch_model.py
@@ -147,7 +147,7 @@ def __init__(
         self.anomaly_map_generator = AnomalyMapGenerator()
 
         self.gaussian = MultiVariateGaussian()
-        self.memory_bank = torch.Tensor()
+        self.memory_bank = torch.empty(0)
 
     def forward(self, input_tensor: torch.Tensor) -> torch.Tensor | InferenceBatch:
         """Forward-pass image-batch (N, C, H, W) into model to extract features.
@@ -224,7 +224,7 @@ def generate_embedding(self, features: dict[str, torch.Tensor]) -> torch.Tensor:
         idx = self.idx.to(embeddings.device)
         return torch.index_select(embeddings, 1, idx)
 
-    def fit_gaussian(self) -> None:
+    def fit(self) -> None:
         """Fits a Gaussian model to the current contents of the memory bank.
 
         This method is typically called after the memory bank has been filled during training.
@@ -242,4 +242,4 @@ def fit_gaussian(self) -> None:
         self.gaussian.fit(self.memory_bank)
 
         # clear memory bank, redcues gpu usage
-        self.memory_bank = torch.Tensor().to(self.memory_bank)
+        self.memory_bank = torch.empty(0).to(self.memory_bank)
diff --git a/src/anomalib/models/image/patchcore/torch_model.py b/src/anomalib/models/image/patchcore/torch_model.py
index 175e06baf0..1970e6dcc2 100644
--- a/src/anomalib/models/image/patchcore/torch_model.py
+++ b/src/anomalib/models/image/patchcore/torch_model.py
@@ -33,7 +33,6 @@
         Coreset subsampling using k-center-greedy approach
 """
 
-import warnings
 from collections.abc import Sequence
 from typing import TYPE_CHECKING
 
@@ -43,6 +42,7 @@
 
 from anomalib.data import InferenceBatch
 from anomalib.models.components import DynamicBufferMixin, KCenterGreedy, TimmFeatureExtractor
+from anomalib.utils import deprecate
 
 from .anomaly_map import AnomalyMapGenerator
 
@@ -125,7 +125,7 @@ def __init__(
         self.feature_pooler = torch.nn.AvgPool2d(3, 1, 1)
         self.anomaly_map_generator = AnomalyMapGenerator()
         self.memory_bank: torch.Tensor
-        self.register_buffer("memory_bank", torch.Tensor())
+        self.register_buffer("memory_bank", torch.empty(0))
 
     def forward(self, input_tensor: torch.Tensor) -> torch.Tensor | InferenceBatch:
         """Process input tensor through the model.
@@ -250,6 +250,7 @@ def reshape_embedding(embedding: torch.Tensor) -> torch.Tensor:
         embedding_size = embedding.size(1)
         return embedding.permute(0, 2, 3, 1).reshape(-1, embedding_size)
 
+    @deprecate(args={"embeddings": None}, since="2.1.0", reason="Use the default memory bank instead.")
     def subsample_embedding(self, sampling_ratio: float, embeddings: torch.Tensor = None) -> None:
         """Subsample the memory_banks embeddings using coreset selection.
 
@@ -269,12 +270,8 @@ def subsample_embedding(self, sampling_ratio: float, embeddings: torch.Tensor =
             torch.Size([100, 512])
         """
         if embeddings is not None:
-            warnings.warn(
-                "The 'embeddings' argument is deprecated and will be removed in a future release."
-                "Please use the default memory bank instead.",
-                DeprecationWarning,
-                stacklevel=2,
-            )
+            del embeddings
+
         if self.memory_bank.size(0) == 0:
             msg = "Memory bank is empty. Cannot perform coreset selection."
             raise ValueError(msg)
diff --git a/src/anomalib/utils/deprecation.py b/src/anomalib/utils/deprecation.py
index baa91030ce..0f195bea8d 100644
--- a/src/anomalib/utils/deprecation.py
+++ b/src/anomalib/utils/deprecation.py
@@ -109,7 +109,7 @@ def _build_deprecation_msg(
 
 
 def _warn_deprecated_arguments(
-    args_map: dict[str, str],
+    args_map: dict[str, str | None],
     bound_args: inspect.BoundArguments,
     since: str | None,
     remove: str | None,
@@ -143,7 +143,7 @@ def _wrap_deprecated_init(
     remove: str | None,
     use: str | None,
     reason: str | None,
-    args_map: dict[str, str] | None,
+    args_map: dict[str, str | None] | None,
     warning_category: type[Warning],
 ) -> type:
     """Wrap a class constructor to emit a deprecation warning when instantiated.
@@ -186,7 +186,7 @@ def _wrap_deprecated_function(
     remove: str | None,
     use: str | None,
     reason: str | None,
-    args_map: dict[str, str] | None,
+    args_map: dict[str, str | None] | None,
     warning_category: type[Warning],
 ) -> Callable:
     """Wrap a function to emit a deprecation warning when called.
@@ -237,7 +237,7 @@ def deprecate(
     remove: str | None = None,
     use: str | None = None,
     reason: str | None = None,
-    args: dict[str, str] | None = None,
+    args: dict[str, str | None] | None = None,
     warning_category: type[Warning] = FutureWarning,
 ) -> Callable[[_T], _T]: ...
 
@@ -249,7 +249,7 @@ def deprecate(
     remove: str | None = None,
     use: str | None = None,
     reason: str | None = None,
-    args: dict[str, str] | None = None,
+    args: dict[str, str | None] | None = None,
     warning_category: type[Warning] = DeprecationWarning,
 ) -> Any:
     """Mark a function, class, or keyword argument as deprecated.
diff --git a/tests/unit/utils/test_deprecate.py b/tests/unit/utils/test_deprecate.py
index 5b4e98589f..936e8d03c3 100644
--- a/tests/unit/utils/test_deprecate.py
+++ b/tests/unit/utils/test_deprecate.py
@@ -32,6 +32,17 @@ def __init__(self) -> None:
     assert instance.val == 42
 
 
+def test_deprecated_function_arg_warns_with_none_replacement() -> None:
+    """Test deprecated argument triggers warning when replacement is None."""
+
+    @deprecate(args={"old_param": None}, since="1.0", remove="2.0")
+    def func(new_param: int | None = None, old_param: int | None = None) -> int:
+        return new_param if new_param is not None else old_param
+
+    with pytest.warns(DeprecationWarning, match=r"(?=.*old_param)(?=.*1\.0)(?=.*2\.0)"):
+        assert func(old_param=10) == 10
+
+
 def test_deprecated_function_arg_warns() -> None:
     """Test deprecated argument triggers warning with correct match."""
 

From f123996141ad7256f0d870ebfddb45a57de9ea4b Mon Sep 17 00:00:00 2001
From: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
Date: Tue, 15 Jul 2025 12:29:41 +0100
Subject: [PATCH 16/16] add type hint for dfkde

Signed-off-by: Alfie Roddan <51797647+alfieroddan@users.noreply.github.com>
---
 src/anomalib/models/image/dfkde/lightning_model.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/anomalib/models/image/dfkde/lightning_model.py b/src/anomalib/models/image/dfkde/lightning_model.py
index 99ac990e1d..4222d626b3 100644
--- a/src/anomalib/models/image/dfkde/lightning_model.py
+++ b/src/anomalib/models/image/dfkde/lightning_model.py
@@ -104,7 +104,7 @@ def __init__(
             visualizer=visualizer,
         )
 
-        self.model = DfkdeModel(
+        self.model: DfkdeModel = DfkdeModel(
             layers=layers,
             backbone=backbone,
             pre_trained=pre_trained,