Merge v1.0.1 to main (#1905)

* Update CHANGELOG.md (#1899)

* Update __init__.py

* Fix metric bug (#1900)

* check metrics by name instead of key

* remove prefix from temp metric collection

* 🔒 Bump onnx version as the previous version caused security issues (#1907)

* 📚 Add action to check broken links (#1838)

* Fix broken links

* Add link checker

* Fix link in template

* linter

* nightly check

---------

Co-authored-by: Samet Akcay <samet.akcay@intel.com>

* Use LabelName ENUM for label assignment (#1881)

* Use LabelName ENUM for label assignment

* Update src/anomalib/deploy/inferencers/torch_inferencer.py

Co-authored-by: Samet Akcay <samet.akcay@intel.com>

---------

Co-authored-by: Samet Akcay <samet.akcay@intel.com>

* Changed prediction label assignment for OpenVINO inferencer (#1872)

* Update openvino_inferencer.py

Fixed incorrect label assignment.
Was True/False, now it's "Anomalous/Normal" as for the PyTorch inferencer in torch_inferencer.py

* Fix ruff

---------

Co-authored-by: Samet <samet.akcay@intel.com>

* Bump onnx version as the previous version caused security issues

---------

Co-authored-by: Alexander Dokuchaev <alexander.dokuchaev@intel.com>
Co-authored-by: TurboJonte <turbojonte@gmail.com>

* Add: fix image transform to float (#1902)

* 📚 Add action to check broken links (#1838)

* Fix broken links

* Add link checker

* Fix link in template

* linter

* nightly check

---------

Co-authored-by: Samet Akcay <samet.akcay@intel.com>

* Use LabelName ENUM for label assignment (#1881)

* Use LabelName ENUM for label assignment

* Update src/anomalib/deploy/inferencers/torch_inferencer.py

Co-authored-by: Samet Akcay <samet.akcay@intel.com>

---------

Co-authored-by: Samet Akcay <samet.akcay@intel.com>

* Changed prediction label assignment for OpenVINO inferencer (#1872)

* Update openvino_inferencer.py

Fixed incorrect label assignment.
Was True/False, now it's "Anomalous/Normal" as for the PyTorch inferencer in torch_inferencer.py

* Fix ruff

---------

Co-authored-by: Samet <samet.akcay@intel.com>

* Add: fix image transform to float

Signed-off-by: Bepitic <bepitic@gmail.com>

* upd: video dtpe transform from torchvision

Signed-off-by: Bepitic <bepitic@gmail.com>

* Added the scale.

Signed-off-by: Bepitic <bepitic@gmail.com>

* Update src/anomalib/data/base/video.py

Co-authored-by: Dick Ameln <amelndjd@gmail.com>

---------

Signed-off-by: Bepitic <bepitic@gmail.com>
Co-authored-by: Alexander Dokuchaev <alexander.dokuchaev@intel.com>
Co-authored-by: Samet Akcay <samet.akcay@intel.com>
Co-authored-by: TurboJonte <turbojonte@gmail.com>
Co-authored-by: Dick Ameln <amelndjd@gmail.com>

* Fix video bug and add tests (#1910)

* 📚 Add action to check broken links (#1838)

* Fix broken links

* Add link checker

* Fix link in template

* linter

* nightly check

---------

Co-authored-by: Samet Akcay <samet.akcay@intel.com>

* squeeze temporal dimension when clip length is 1

* add unit test for video input tensor dtype

* add test for single frame clips

* update changelog

* add author

---------

Co-authored-by: Alexander Dokuchaev <alexander.dokuchaev@intel.com>
Co-authored-by: Samet Akcay <samet.akcay@intel.com>

* Temporarily set devices to 1

* 🐞 Add greyscale conversion (#1911)

Add greyscale conversion

Signed-off-by: Ashwin Vaidya <ashwinnitinvaidya@gmail.com>

* 🐞 Fix publish workflow (#1912)

Disable tty prompt for gpg import

Signed-off-by: Ashwin Vaidya <ashwinnitinvaidya@gmail.com>

---------

Signed-off-by: Bepitic <bepitic@gmail.com>
Signed-off-by: Ashwin Vaidya <ashwinnitinvaidya@gmail.com>
Co-authored-by: Dick Ameln <dick.ameln@intel.com>
Co-authored-by: Alexander Dokuchaev <alexander.dokuchaev@intel.com>
Co-authored-by: TurboJonte <turbojonte@gmail.com>
Co-authored-by: Paco <bepitic@gmail.com>
Co-authored-by: Dick Ameln <amelndjd@gmail.com>
Co-authored-by: Ashwin Vaidya <ashwin.vaidya@intel.com>

.github/workflows/publish.yml

@@ -9,15 +9,17 @@ jobs:
     runs-on: ubuntu-20.04
     steps:
       - uses: actions/checkout@v2
-      - uses: actions/setup-python@v2
+      - uses: actions/setup-python@v4
+        with:
+          python-version: "3.10"
       - name: Install dependencies
         run: |
           python -m pip install --upgrade pip
           pip install setuptools wheel twine
 
       - name: Import GPG Key
         run: |
-          echo "${{ secrets.GPG_PRIVATE_KEY }}" | base64 --decode | gpg --import
+          echo "${{ secrets.GPG_PRIVATE_KEY }}" | base64 --decode | gpg --import --batch --yes --no-tty
 
       - name: Build and sign distribution
         run: |

CHANGELOG.md

@@ -14,9 +14,49 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
 
 ### Fixed
 
-- Use right interpolation method in WinCLIP resize (<https://github.com/openvinotoolkit/anomalib/pull/1889>)
+### New Contributors
+
+**Full Changelog**:
+
+## [v1.0.1] - Unreleased
+
+### Added
+
+- Add requirements into `pyproject.toml` & Refactor anomalib install `get_requirements` by @harimkang in https://github.com/openvinotoolkit/anomalib/pull/1808
+
+### Changed
+
+- 📚 Update the getting started notebook by @samet-akcay in https://github.com/openvinotoolkit/anomalib/pull/1800
+- 🔨Refactored-assert-statements-with-explicit-error-handling by @sahusiddharth in https://github.com/openvinotoolkit/anomalib/pull/1825
+- 🔨Made-imagenette-path-configurable-in-config by @sahusiddharth in https://github.com/openvinotoolkit/anomalib/pull/1833
+- 🛠️ Update changelog by @ashwinvaidya17 in https://github.com/openvinotoolkit/anomalib/pull/1842
+- Remove input_size argument from models #1827 by @Shakib-IO in https://github.com/openvinotoolkit/anomalib/pull/1856
+- 🚀 Allow validation splits from training data by @davnn in https://github.com/openvinotoolkit/anomalib/pull/1865
+- 🛠️ Ensure images are loaded in RGB format by @davnn in https://github.com/openvinotoolkit/anomalib/pull/1866
+- 🔨 Update OpenVINO predict to handle normalization inside the method. by @samet-akcay in https://github.com/openvinotoolkit/anomalib/pull/1875
+- ✨ Upgrade TorchMetrics by @ashwinvaidya17 in https://github.com/openvinotoolkit/anomalib/pull/1879
+- Address minor WinCLIP issues by @djdameln in https://github.com/openvinotoolkit/anomalib/pull/1889
+
+### Deprecated
+
+### Fixed
+
+- 🐞 Fix single-frame video input size by [@djdameln](https://github.com/djdameln) (<https://github.com/openvinotoolkit/anomalib/pull/1910>)
+- 🐞 Fix dobot notebook by @djdameln in https://github.com/openvinotoolkit/anomalib/pull/1852
+- 🐞 Fix CLI config and update the docs. by @samet-akcay in https://github.com/openvinotoolkit/anomalib/pull/1886
+- 🐞 Fix the error if the device in masks_to_boxes is not both CPU and CUDA by @danylo-boiko in https://github.com/openvinotoolkit/anomalib/pull/1839
+- 🐞 Hot-fix wrong requirement for setup.py by @harimkang in https://github.com/openvinotoolkit/anomalib/pull/1823
+- 🐞 Use right interpolation method in WinCLIP resize (<https://github.com/openvinotoolkit/anomalib/pull/1889>)
 - 🐞 Fix the error if the device in masks_to_boxes is not both CPU and CUDA by @danylo-boiko in https://github.com/openvinotoolkit/anomalib/pull/1839
 
+### New Contributors
+
+- @sahusiddharth made their first contribution in https://github.com/openvinotoolkit/anomalib/pull/1825
+- @Shakib-IO made their first contribution in https://github.com/openvinotoolkit/anomalib/pull/1856
+- @davnn made their first contribution in https://github.com/openvinotoolkit/anomalib/pull/1866
+
+**Full Changelog**: https://github.com/openvinotoolkit/anomalib/compare/v1.0.0...v1.0.1
+
 ## [v1.0.0] - 2024-02-29
 
 ### Added

pyproject.toml

@@ -39,7 +39,7 @@ core = [
     "torchmetrics==0.10.3",
     "open-clip-torch>=2.23.0",
 ]
-openvino = ["openvino-dev>=2023.0", "nncf>=2.5.0", "onnx>=1.13.1"]
+openvino = ["openvino-dev>=2023.0", "nncf>=2.5.0", "onnx>=1.16.0"]
 loggers = [
     "comet-ml>=3.31.7",
     "gradio>=4",

requirements/openvino.txt

@@ -1,3 +1,3 @@
 openvino-dev>=2023.0
 nncf>=2.5.0
-onnx>=1.13.1
+onnx>=1.16.0

src/anomalib/__init__.py

@@ -5,7 +5,7 @@
 
 from enum import Enum
 
-__version__ = "1.1.0dev"
+__version__ = "1.0.1"
 
 
 class LearningType(str, Enum):

src/anomalib/callbacks/metrics.py

@@ -91,14 +91,9 @@ def setup(
         if isinstance(pl_module, AnomalyModule):
             pl_module.image_metrics = create_metric_collection(image_metric_names, "image_")
             if hasattr(pl_module, "pixel_metrics"):  # incase metrics are loaded from model checkpoint
-                new_metrics = create_metric_collection(pixel_metric_names, "pixel_")
+                new_metrics = create_metric_collection(pixel_metric_names)
                 for name in new_metrics:
-                    # Ruff is ignored as
-                    # >>> name not in pl_module.pixel_metrics
-                    # True
-                    # >>> name not in pl_module.pixel_metrics.keys()
-                    # False
-                    if name not in pl_module.pixel_metrics.keys():  # noqa: SIM118
+                    if name not in pl_module.pixel_metrics:
                         pl_module.pixel_metrics.add_metrics(new_metrics[name])
             else:
                 pl_module.pixel_metrics = create_metric_collection(pixel_metric_names, "pixel_")

src/anomalib/data/base/video.py

@@ -10,6 +10,7 @@
 import torch
 from pandas import DataFrame
 from torchvision.transforms.v2 import Transform
+from torchvision.transforms.v2.functional import to_dtype_video
 from torchvision.tv_tensors import Mask
 
 from anomalib import TaskType
@@ -153,6 +154,7 @@ def __getitem__(self, index: int) -> dict[str, str | torch.Tensor]:
             msg = "self.indexer must be an instance of ClipsIndexer."
             raise TypeError(msg)
         item = self.indexer.get_item(index)
+        item["image"] = to_dtype_video(video=item["image"], scale=True)
         # include the untransformed image for visualization
         item["original_image"] = item["image"].to(torch.uint8)
 
@@ -167,6 +169,9 @@ def __getitem__(self, index: int) -> dict[str, str | torch.Tensor]:
         elif self.transform:
             item["image"] = self.transform(item["image"])
 
+        # squeeze temporal dimensions in case clip length is 1
+        item["image"] = item["image"].squeeze(0)
+
         # include only target frame in gt
         if self.clip_length_in_frames > 1 and self.target_frame != VideoTargetFrame.ALL:
             item = self._select_targets(item)

src/anomalib/data/utils/image.py

@@ -371,7 +371,7 @@ def read_mask(path: str | Path, as_tensor: bool = False) -> torch.Tensor | np.nd
         >>> type(mask)
         <class 'torch.Tensor'>
     """
-    image = Image.open(path)
+    image = Image.open(path).convert("L")
     return Mask(to_image(image).squeeze() / 255, dtype=torch.uint8) if as_tensor else np.array(image)
 
 

src/anomalib/engine/engine.py

@@ -301,6 +301,9 @@ def _setup_trainer(self, model: AnomalyModule) -> None:
         # Setup anomalib callbacks to be used with the trainer
         self._setup_anomalib_callbacks()
 
+        # Temporarily set devices to 1 to avoid issues with multiple processes
+        self._cache.args["devices"] = 1
+
         # Instantiate the trainer if it is not already instantiated
         if self._trainer is None:
             self._trainer = Trainer(**self._cache.args)

src/anomalib/metrics/__init__.py

@@ -162,7 +162,7 @@ def metric_collection_from_dicts(metrics: dict[str, dict[str, Any]], prefix: str
 
 def create_metric_collection(
     metrics: list[str] | dict[str, dict[str, Any]],
-    prefix: str | None,
+    prefix: str | None = None,
 ) -> AnomalibMetricCollection:
     """Create a metric collection from a list of metric names or dictionaries.
 

tests/unit/data/base/video.py

@@ -3,8 +3,8 @@
 # Copyright (C) 2023-2024 Intel Corporation
 # SPDX-License-Identifier: Apache-2.0
 
-
 import pytest
+import torch
 
 from anomalib.data import AnomalibDataModule
 
@@ -46,3 +46,27 @@ def test_get_item_returns_correct_keys_and_shapes(self, datamodule: AnomalibData
             assert len(batch["label"]) == 4
             assert batch["mask"].shape == (4, 256, 256)
             assert batch["mask"].shape == (4, 256, 256)
+
+    @pytest.mark.parametrize("subset", ["train", "val", "test"])
+    def test_item_dtype(self, datamodule: AnomalibDataModule, subset: str) -> None:
+        """Test that the input tensor is of float type and scaled between 0-1."""
+        # Get the dataloader.
+        dataloader = getattr(datamodule, f"{subset}_dataloader")()
+
+        # Get the first batch.
+        batch = next(iter(dataloader))
+        clip = batch["image"]
+        assert clip.dtype == torch.float32
+        assert clip.min() >= 0
+        assert clip.max() <= 1
+
+    @pytest.mark.parametrize("clip_length_in_frames", [1])
+    def test_single_frame_squeezed(self, datamodule: AnomalibDataModule) -> None:
+        """Test that the temporal dimension is squeezed when the clip lenght is 1."""
+        # Get the dataloader.
+        dataloader = datamodule.train_dataloader()
+
+        # Get the first batch.
+        batch = next(iter(dataloader))
+        clip = batch["image"]
+        assert clip.shape == (4, 3, 256, 256)

tests/unit/data/video/test_avenue.py

@@ -16,11 +16,17 @@ class TestAvenue(_TestAnomalibVideoDatamodule):
     """Avenue Datamodule Unit Tests."""
 
     @pytest.fixture()
-    def datamodule(self, dataset_path: Path, task_type: TaskType) -> Avenue:
+    def clip_length_in_frames(self) -> int:
+        """Return the number of frames in each clip."""
+        return 2
+
+    @pytest.fixture()
+    def datamodule(self, dataset_path: Path, task_type: TaskType, clip_length_in_frames: int) -> Avenue:
         """Create and return a Avenue datamodule."""
         _datamodule = Avenue(
             root=dataset_path / "avenue",
             gt_dir=dataset_path / "avenue" / "ground_truth_demo",
+            clip_length_in_frames=clip_length_in_frames,
             image_size=256,
             task=task_type,
             num_workers=0,

tests/unit/data/video/test_shanghaitech.py

@@ -17,11 +17,17 @@ class TestShanghaiTech(_TestAnomalibVideoDatamodule):
     """ShanghaiTech Datamodule Unit Tests."""
 
     @pytest.fixture()
-    def datamodule(self, dataset_path: Path, task_type: TaskType) -> ShanghaiTech:
+    def clip_length_in_frames(self) -> int:
+        """Return the number of frames in each clip."""
+        return 2
+
+    @pytest.fixture()
+    def datamodule(self, dataset_path: Path, task_type: TaskType, clip_length_in_frames: int) -> ShanghaiTech:
         """Create and return a Shanghai datamodule."""
         _datamodule = ShanghaiTech(
             root=dataset_path / "shanghaitech",
             scene=1,
+            clip_length_in_frames=clip_length_in_frames,
             image_size=(256, 256),
             train_batch_size=4,
             eval_batch_size=4,

tests/unit/data/video/test_ucsdped.py

@@ -16,11 +16,17 @@ class TestUCSDped(_TestAnomalibVideoDatamodule):
     """UCSDped Datamodule Unit Tests."""
 
     @pytest.fixture()
-    def datamodule(self, dataset_path: Path, task_type: TaskType) -> UCSDped:
+    def clip_length_in_frames(self) -> int:
+        """Return the number of frames in each clip."""
+        return 2
+
+    @pytest.fixture()
+    def datamodule(self, dataset_path: Path, task_type: TaskType, clip_length_in_frames: int) -> UCSDped:
         """Create and return a UCSDped datamodule."""
         _datamodule = UCSDped(
             root=dataset_path / "ucsdped",
             category="dummy",
+            clip_length_in_frames=clip_length_in_frames,
             task=task_type,
             image_size=256,
             train_batch_size=4,