Merge main into feature/pimo (#1273)

* Configure readthedocs via `.readthedocs.yaml` file (#1229)

* 🚚 Refactor Benchmarking Script (#1216)

* New printing stuff

* Remove dead code + address codacy issues

* Refactor try/except + log to comet/wandb during runs

* pre-commit error

* third-party configuration

---------

Co-authored-by: Ashwin Vaidya <ashwinitinvaidya@gmail.com>

* Update CODEOWNERS

* Enable training with only normal images for MVTec (#1241)

* ignore mask check when dataset has only normal samples

* update changelog

* Revert "🚚 Refactor Benchmarking Script" (#1239)

Revert "🚚 Refactor Benchmarking Script (#1216)"

This reverts commit 784767f.

* Update benchmarking notebook (#1242)

* Fix metadata path

* Update benchmarking notebook

* Fix links to model architecture images (#1245)

* Fix links to architecture images

* Change links to raw files

* Wandb unwatch method belongs to experiment, not logger (#1246)

unwatch method belongs to experiment, not logger

* (Minor change) Added the tracer_kwargs to the TorchFXFeatureExtractor class (#1214)

* Added tracer_kwargs to torchfx

* Added tracer_kwargs on docstring

* Replace cdist in Patchcore (#1267)

* Ignore hidden directories when creating `Folder` dataset (#1268)

* Remove `config` from argparse in OpenVINO inference script. (#1257)

* Fix metadata path

* Remove leftover argument

* Update openvino entrypoint script

* Fix EfficientAD number of steps for optimizer lr change? (#1266)

* Fix metadata path

* Fix number of steps

---------

Co-authored-by: Ashwin Vaidya <ashwin.vaidya@intel.com>
Co-authored-by: Ashwin Vaidya <ashwinitinvaidya@gmail.com>
Co-authored-by: Dick Ameln <dick.ameln@intel.com>
Co-authored-by: Blaž Rolih <61357777+blaz-r@users.noreply.github.com>
Co-authored-by: Sean Aubin <seanaubin@gmail.com>
Co-authored-by: JoaoGuibs <32060480+JoaoGuibs@users.noreply.github.com>

src/anomalib/data/utils/path.py

@@ -58,7 +58,11 @@ def _prepare_files_labels(
     if isinstance(extensions, str):
         extensions = (extensions,)
 
-    filenames = [f for f in path.glob(r"**/*") if f.suffix in extensions and not f.is_dir()]
+    filenames = [
+        f
+        for f in path.glob("**/*")
+        if f.suffix in extensions and not f.is_dir() and not any(part.startswith(".") for part in f.parts)
+    ]
     if not filenames:
         raise RuntimeError(f"Found 0 {path_type} images in {path}")
 

src/anomalib/models/cfa/README.md

@@ -12,7 +12,7 @@ Coupled-hypersphere-based Feature Adaptation (CFA) localizes anomalies using fea
 
 ## Architecture
 
-![Cfa Architecture](../../../docs/source/images/cfa/architecture.png "Cfa Architecture")
+![Cfa Architecture](https://raw.githubusercontent.com/openvinotoolkit/anomalib/main/docs/source/images/cfa/architecture.png "Cfa Architecture")
 
 ## Usage
 

src/anomalib/models/cflow/README.md

@@ -10,7 +10,7 @@ CFLOW model is based on a conditional normalizing flow framework adopted for ano
 
 ## Architecture
 
-![CFlow Architecture](../../../docs/source/images/cflow/architecture.jpg "CFlow Architecture")
+![CFlow Architecture](https://raw.githubusercontent.com/openvinotoolkit/anomalib/main/docs/source/images/cflow/architecture.jpg "CFlow Architecture")
 
 ## Usage
 

src/anomalib/models/components/feature_extractors/torchfx.py

@@ -39,6 +39,9 @@ class TorchFXFeatureExtractor(nn.Module):
             path for custom models.
         requires_grad (bool): Models like ``stfpm`` use the feature extractor for training. In such cases we should
             set ``requires_grad`` to ``True``. Default is ``False``.
+        tracer_kwargs (dict | None): a dictionary of keyword arguments for NodePathTracer (which passes them onto
+            it's parent class torch.fx.Tracer). Can be used to allow not tracing through a list of problematic
+            modules, by passing a list of `leaf_modules` as one of the `tracer_kwargs`.
 
     Example:
         With torchvision models:
@@ -91,6 +94,7 @@ def __init__(
         return_nodes: list[str],
         weights: str | WeightsEnum | None = None,
         requires_grad: bool = False,
+        tracer_kwargs: dict | None = None,
     ):
         super().__init__()
         if isinstance(backbone, dict):
@@ -102,14 +106,17 @@ def __init__(
                 f"backbone needs to be of type str | BackboneParams | dict | nn.Module, but was type {type(backbone)}"
             )
 
-        self.feature_extractor = self.initialize_feature_extractor(backbone, return_nodes, weights, requires_grad)
+        self.feature_extractor = self.initialize_feature_extractor(
+            backbone, return_nodes, weights, requires_grad, tracer_kwargs
+        )
 
     def initialize_feature_extractor(
         self,
         backbone: BackboneParams | nn.Module,
         return_nodes: list[str],
         weights: str | WeightsEnum | None = None,
         requires_grad: bool = False,
+        tracer_kwargs: dict | None = None,
     ) -> GraphModule:
         """Extract features from a CNN.
 
@@ -125,6 +132,9 @@ class can be provided and it will try to load the weights from the provided weig
                 path for custom models.
             requires_grad (bool): Models like ``stfpm`` use the feature extractor for training. In such cases we should
                 set ``requires_grad`` to ``True``. Default is ``False``.
+            tracer_kwargs (dict | None): a dictionary of keyword arguments for NodePathTracer (which passes them onto
+                it's parent class torch.fx.Tracer). Can be used to allow not tracing through a list of problematic
+                modules, by passing a list of `leaf_modules` as one of the `tracer_kwargs`.
 
         Returns:
             Feature Extractor based on TorchFX.
@@ -148,7 +158,7 @@ class can be provided and it will try to load the weights from the provided weig
                         model_weights = model_weights["state_dict"]
                     backbone_model.load_state_dict(model_weights)
 
-        feature_extractor = create_feature_extractor(backbone_model, return_nodes)
+        feature_extractor = create_feature_extractor(backbone_model, return_nodes, tracer_kwargs=tracer_kwargs)
 
         if not requires_grad:
             feature_extractor.eval()

src/anomalib/models/csflow/README.md

@@ -23,11 +23,11 @@ The anomaly score for each local position $(i,j)$ of the feature map $y^s$ at sc
 
 ## Architecture
 
-![CS-Flow Architecture](../../../docs/source/images/cs_flow/architecture1.jpg "CS-Flow Architecture")
+![CS-Flow Architecture](https://raw.githubusercontent.com/openvinotoolkit/anomalib/main/docs/source/images/cs_flow/architecture1.jpg "CS-Flow Architecture")
 
-![Architecture of a Coupling Block](../../../docs/source/images/cs_flow/architecture2.jpg "Architecture of a Coupling Block")
+![Architecture of a Coupling Block](https://raw.githubusercontent.com/openvinotoolkit/anomalib/main/docs/source/images/cs_flow/architecture2.jpg "Architecture of a Coupling Block")
 
-![Architecture of network predicting scale and shift parameters.](../../../docs/source/images/cs_flow/architecture3.jpg "Architecture of network predicting scale and shift parameters.")
+![Architecture of network predicting scale and shift parameters.](https://raw.githubusercontent.com/openvinotoolkit/anomalib/main/docs/source/images/cs_flow/architecture3.jpg "Architecture of network predicting scale and shift parameters.")
 
 ## Usage
 

src/anomalib/models/draem/README.md

@@ -12,7 +12,7 @@ For optimal results, DRAEM requires specifying the path to a folder of image dat
 
 ## Architecture
 
-![DRAEM Architecture](../../../docs/source/images/draem/architecture.png "DRAEM Architecture")
+![DRAEM Architecture](https://raw.githubusercontent.com/openvinotoolkit/anomalib/main/docs/source/images/draem/architecture.png "DRAEM Architecture")
 
 ## Usage
 

src/anomalib/models/efficient_ad/lightning_model.py

@@ -204,7 +204,7 @@ def configure_optimizers(self) -> optim.Optimizer:
             lr=self.lr,
             weight_decay=self.weight_decay,
         )
-        num_steps = max(
+        num_steps = min(
             self.trainer.max_steps, self.trainer.max_epochs * len(self.trainer.datamodule.train_dataloader())
         )
         scheduler = optim.lr_scheduler.StepLR(optimizer, step_size=int(0.95 * num_steps), gamma=0.1)

src/anomalib/models/fastflow/README.md

@@ -10,7 +10,7 @@ FastFlow is a two-dimensional normalizing flow-based probability distribution es
 
 ## Architecture
 
-![FastFlow Architecture](../../../docs/source/images/fastflow/architecture.jpg "FastFlow Architecture")
+![FastFlow Architecture](https://raw.githubusercontent.com/openvinotoolkit/anomalib/main/docs/source/images/fastflow/architecture.jpg "FastFlow Architecture")
 
 ## Usage
 

src/anomalib/models/ganomaly/README.md

@@ -12,7 +12,7 @@ The key idea here is that, during inference, when an anomalous image is passed t
 
 ## Architecture
 
-![GANomaly Architecture](../../../docs/source/images/ganomaly/architecture.jpg "GANomaly Architecture")
+![GANomaly Architecture](https://raw.githubusercontent.com/openvinotoolkit/anomalib/main/docs/source/images/ganomaly/architecture.jpg "GANomaly Architecture")
 
 ## Usage
 

src/anomalib/models/padim/README.md

@@ -12,7 +12,7 @@ During inference, Mahalanobis distance is used to score each patch position of t
 
 ## Architecture
 
-![PaDiM Architecture](../../../docs/source/images/padim/architecture.jpg "PaDiM Architecture")
+![PaDiM Architecture](https://raw.githubusercontent.com/openvinotoolkit/anomalib/main/docs/source/images/padim/architecture.jpg "PaDiM Architecture")
 
 ## Usage
 

src/anomalib/models/patchcore/README.md

@@ -12,7 +12,7 @@ During inference this memory bank is coreset subsampled. Coreset subsampling gen
 
 ## Architecture
 
-![PatchCore Architecture](../../../../docs/source/images/patchcore/architecture.jpg "PatchCore Architecture")
+![PatchCore Architecture](https://raw.githubusercontent.com/openvinotoolkit/anomalib/main/docs/source/images/patchcore/architecture.jpg "PatchCore Architecture")
 
 ## Usage
 

src/anomalib/models/patchcore/torch_model.py

@@ -142,6 +142,28 @@ def subsample_embedding(self, embedding: Tensor, sampling_ratio: float) -> None:
         coreset = sampler.sample_coreset()
         self.memory_bank = coreset
 
+    @staticmethod
+    def euclidean_dist(x: Tensor, y: Tensor) -> Tensor:
+        """
+        Calculates pair-wise distance between row vectors in x and those in y.
+
+        Replaces torch cdist with p=2, as cdist is not properly exported to onnx and openvino format.
+        Resulting matrix is indexed by x vectors in rows and y vectors in columns.
+
+        Args:
+            x: input tensor 1
+            y: input tensor 2
+
+        Returns:
+            Matrix of distances between row vectors in x and y.
+        """
+        x_norm = x.pow(2).sum(dim=-1, keepdim=True)  # |x|
+        y_norm = y.pow(2).sum(dim=-1, keepdim=True)  # |y|
+        # row distance can be rewritten as sqrt(|x| - 2 * x @ y.T + |y|.T)
+        res = x_norm - 2 * torch.matmul(x, y.transpose(-2, -1)) + y_norm.transpose(-2, -1)
+        res = res.clamp_min_(0).sqrt_()
+        return res
+
     def nearest_neighbors(self, embedding: Tensor, n_neighbors: int) -> tuple[Tensor, Tensor]:
         """Nearest Neighbours using brute force method and euclidean norm.
 
@@ -153,7 +175,7 @@ def nearest_neighbors(self, embedding: Tensor, n_neighbors: int) -> tuple[Tensor
             Tensor: Patch scores.
             Tensor: Locations of the nearest neighbor(s).
         """
-        distances = torch.cdist(embedding, self.memory_bank, p=2.0)  # euclidean norm
+        distances = self.euclidean_dist(embedding, self.memory_bank)
         if n_neighbors == 1:
             # when n_neighbors is 1, speed up computation by using min instead of topk
             patch_scores, locations = distances.min(1)
@@ -188,7 +210,7 @@ def compute_anomaly_score(self, patch_scores: Tensor, locations: Tensor, embeddi
         # indices of N_b(m^*) in the paper
         _, support_samples = self.nearest_neighbors(nn_sample, n_neighbors=self.num_neighbors)
         # 4. Find the distance of the patch features to each of the support samples
-        distances = torch.cdist(max_patches_features.unsqueeze(1), self.memory_bank[support_samples], p=2.0)
+        distances = self.euclidean_dist(max_patches_features.unsqueeze(1), self.memory_bank[support_samples])
         # 5. Apply softmax to find the weights
         weights = (1 - F.softmax(distances.squeeze(1), 1))[..., 0]
         # 6. Apply the weight factor to the score

src/anomalib/models/reverse_distillation/README.md

@@ -12,7 +12,7 @@ During testing, a similar step is followed but this time the cosine distance bet
 
 ## Architecture
 
-![Anomaly Detection via Reverse Distillation from One-Class Embedding Architecture](../../../docs/source/images/reverse_distillation/architecture.png "Reverse Distillation Architecture")
+![Anomaly Detection via Reverse Distillation from One-Class Embedding Architecture](https://raw.githubusercontent.com/openvinotoolkit/anomalib/main/docs/source/images/reverse_distillation/architecture.png "Reverse Distillation Architecture")
 
 ## Usage
 

src/anomalib/models/stfpm/README.md

@@ -12,7 +12,7 @@ During inference, the feature pyramids of teacher and student networks are compa
 
 ## Architecture
 
-![STFPM Architecture](../../../docs/source/images/stfpm/architecture.jpg "STFPM Architecture")
+![STFPM Architecture](https://raw.githubusercontent.com/openvinotoolkit/anomalib/main/docs/source/images/stfpm/architecture.jpg "STFPM Architecture")
 
 ## Usage
 

src/anomalib/utils/callbacks/graph.py

@@ -43,4 +43,4 @@ def on_train_end(self, trainer: Trainer, pl_module: LightningModule) -> None:
             if isinstance(logger, (AnomalibCometLogger, AnomalibTensorBoardLogger)):
                 logger.log_graph(pl_module, input_array=torch.ones((1, 3, 256, 256)))
             elif isinstance(logger, AnomalibWandbLogger):
-                logger.unwatch(pl_module)  # type: ignore
+                logger.experiment.unwatch(pl_module)  # type: ignore

tests/pre_merge/tools/test_openvino_entrypoint.py

@@ -51,8 +51,6 @@ def test_openvino_inference(
 
         arguments = get_parser().parse_args(
             [
-                "--config",
-                "src/anomalib/models/padim/config.yaml",
                 "--weights",
                 project_path + "/weights/openvino/model.bin",
                 "--metadata",

tools/inference/openvino_inference.py

@@ -1,7 +1,7 @@
-"""Anomalib Inferencer Script.
+"""Anomalib OpenVINO Inferencer Script.
 
-This script performs inference by reading a model config file from
-command line, and show the visualization results.
+This script performs OpenVINO inference by reading a model from
+file system, and show the visualization results.
 """
 
 # Copyright (C) 2022 Intel Corporation
@@ -23,7 +23,6 @@ def get_parser() -> ArgumentParser:
         ArgumentParser: The parser object.
     """
     parser = ArgumentParser()
-    parser.add_argument("--config", type=Path, required=True, help="Path to a config file")
     parser.add_argument("--weights", type=Path, required=True, help="Path to model weights")
     parser.add_argument("--metadata", type=Path, required=True, help="Path to a JSON file containing the metadata.")
     parser.add_argument("--input", type=Path, required=True, help="Path to an image to infer.")