Hazelnut Toy Dataset

This relase contains the hazelnut toy dataset that is used to demonstrate custom data operations that could be done with anomalib. It is a toy dataset for quick demonstration and experimentation.

v0.7.0

What's Changed

[v0.7.0] - 2023-08-28

Added

• AUPRO binning capability by @Yann-CV in #1145
• Add support for receiving dataset paths as a list by @harimkang in #1265
• Add modelAPI compatible OpenVINO export by @ashwinvaidya17 in #1219

Changed

• Enable training with only normal images for MVTec by @djdameln in #1241
• Improve default settings of EfficientAD by @nelson1425 in #1143
• Added the tracer_kwargs to the TorchFXFeatureExtractor class by @JoaoGuibs in #1214
• Replace cdist in Patchcore by @blaz-r in #1267
• Ignore hidden directories when creating Folder dataset by @samet-akcay in #1268

Fixed

• Fix typechecking error for toch.onnx.export by @ORippler in #1159
• Fix benchmarking type error by @blaz-r in #1155
• Fix incorrect shape mismatch between anomaly map and ground truth mask by @alexriedel1 in #1182
• Fix dataset keys in benchmarking notebook by @samet-akcay in #1242
• Remove config from argparse in OpenVINO inference script by @samet-akcay in #1257
• Fix EfficientAD number of steps for optimizer lr change by @samet-akcay in #1266
• Fix unable to read the mas image by @samet-akcay in #1277

New Contributors

• @nelson1425 made their first contribution in #1143
• @Seanny123 made their first contribution in #1246
• @JoaoGuibs made their first contribution in #1214
• @harimkang made their first contribution in #1265
• @Yann-CV made their first contribution in #1145

Full Changelog: v0.6.0...v0.7.0

DSR Pretrained Weights

Description

This release contains the pre-trained weights for the DSR model. The weights are obtained from https://github.com/VitjanZ/DSR_anomaly_detection.

v0.6.0

What's Changed

[v0.6.0] - 2023-06-15

Added

• EfficientAD by @alexriedel1 in #1073
• AI-VAD bbox post-processing by @djdameln in #1103
• Add dataset categories to data modules by @samet-akcay in #1105
• Pass codedov token from environment by @ashwinvaidya17 in #1102

Changed

• OV API2.0 by @paularamo in #1098

Deprecated

• OV API1.0 by @paularamo in #1098

Fixed

• Fix Fastflow ONNX export. by @jasonvanzelm in #1108
• Fix tile import typo by @xbkaishui in #1106
• Fix pre-commit issues caused by the EfficientAD PR by @samet-akcay in #1114
• Bump requests from 2.26.0 to 2.31.0 in /requirements by @dependabot in #1100

New Contributors

• @xbkaishui made their first contribution in #1106
• @jasonvanzelm made their first contribution in #1108

Full Changelog: v0.5.1...v0.6.0

EfficientAD Pretrained Weights

Description

This release contains the pre-trained weights for the small and medium teacher networks of the EfficientAD model. Please refer to the paper for more details.

v0.5.1

What's Changed

Added

• 🧪 Add tests for tools by @ashwinvaidya17 in #1069
• Add kolektor dataset by @Ravindu987 in #983

Changed

• Rename metadata_path to metadata in OpenvinoInferencer in #1101
• 📝 [Notebooks] - Simplify the dobot notebooks. by @samet-akcay in #1084
• Upgrade python to 3.10 by @samet-akcay in #1035
• 📝 [Notebooks] - Install anomalib via pip in the Jupyter Notebooks by @samet-akcay in #1091
• Update code-scan workflow to use Trivy by @yunchu in #1097

Fixed

• Fix init_state_dict bug in wrap_nncf_model in #1101
• Fix mypy pep561 by @WenjingKangIntel in #1088
• 📝 [Notebooks] Fix the broken formatting by @samet-akcay in #1090

Full Changelog: v0.5.0...v0.5.1

v0.5.0

[v0.5.0] - 2023-05-09

Added

• 📚 Add OpenVINO Inference to getting started notebook. by @samet-akcay in #897
• Mvtec 3d by @alexriedel1 in #907
• MVTec 3D and Folder3D by @alexriedel1 in #942
• add reorder=True argument in aupro by @triet1102 in #944
• always reorder inputs when computing AUROC by @djdameln in #945
• always reorder for aupr metric by @triet1102 in #975
• Add README.md files to notebooks directories and its subdirectories (#993)
• Set transformations from the config file by @alexriedel1 in #990
• Add contributors to README.md by @samet-akcay in #995
• Add codeowners file by @samet-akcay in #1019
• Configure reference frame for multi-frame video clips by @djdameln in #1023
• [Algo] Attribute-based Representations for Accurate and Interpretable Video Anomaly Detection by @djdameln in #1040

Changed

• Switch to src layout by @samet-akcay in #921
• Remove config flag from OpenVINOInferencer (#939)
• Add ruff as the main linter by @samet-akcay in #936
• Add a new workflow for code scanning by @yunchu in #940
• Enable bandit scanning by @yunchu in #954
• 🐳 Update Containers and Readme by @ashwinvaidya17 in #952
• Refactor AUPRO metric by @triet1102 in #991
• enable auto-fixing for ruff in pre-commit by @djdameln in #1004
• Refactor strings and ints into enum.Enum by @WenjingKangIntel in #1044
• Modify codecov upload by @ashwinvaidya17 in #1080

Deprecated

• Remove torchvision and torchtext by @samet-akcay in #903
• Remove codacy from ci docs by @ashwinvaidya17 in #924
• Remove config dependency from OpenVINOInferencer by @samet-akcay in #939
• Remove config from torch inferencer by @samet-akcay in #1001

Fixed

• Bugfix code logic to allow for passing of nn.Module to TorchFXFeatureExtractor by @ORippler in #935
• fix broken links to tutorials (ex guides) by @sergiev in #957
• Fixed outdated info in readme by @blaz-r in #969
• Fix ruff isort integration by @samet-akcay in #976
• Fix/samples dataframe annotation by @samet-akcay in #981
• Fixed openvino_inferencer in gradio_inference by @blaz-r in #972
• Fix issue in tutorial by @Ravindu987 in #997
• Fix tarfile vulnerability by @djdameln in #1003
• Cuda 11.4 dockerfile fix by @phcarval in #1021
• Make anomalib PEP 561 compliant for mypy by @WenjingKangIntel in #1038
• [Bug: 839] Crop in SSPCAB implementation by @isaacncz in #1057
• [Bug: 865] datamodule.setup() assertion failed by @isaacncz in #1058
• Fix logger message for test_split_ratio by @ugotsoul in #1071
• Fix notebook readme formatting by @samet-akcay in #1075

New Contributors

• @triet1102 made their first contribution in #944
• @sergiev made their first contribution in #957
• @blaz-r made their first contribution in #969
• @ineiti made their first contribution in #987
• @Ravindu987 made their first contribution in #997
• @phcarval made their first contribution in #1021
• @WenjingKangIntel made their first contribution in #1038
• @isaacncz made their first contribution in #1057
• @ugotsoul made their first contribution in #1071

Full Changelog: v0.4.0...v0.5.0

v0.4.0

What's Changed

Added

• Add Dobot notebook (#928)
• Add ShanghaiTech Campus video anomaly detection dataset (#869)
• Add pyupgrade to pre-commit configs, and refactor based on pyupgrade and refurb (#845)
• Add CFA model implementation (#783)
• Add RKDE model implementation (#821)
• Add Visual Anomaly (VisA) dataset adapter (#824)
• Add Synthetic anomalous dataset for validation and testing (#822)
• Add Detection task type support (#822)
• Add UCSDped and Avenue dataset implementation (#822)
• Add base classes for video dataset and video datamodule (#822)
• Add base classes for image dataset and image dataModule (#822)
• ✨ Add CSFlow model (#657)
• Log loss for existing trainable models (#804)
• Add section for community project (#768)
• ✨ Add torchfx feature extractor (#675)
• Add tiling notebook (#712)
• Add posargs to tox to enable testing a single file (#695)
• Add option to load metrics with kwargs (#688)
• 🐞 Add device flag to TorchInferencer (#601)

Changed

• Bump OpenVINO version to 2022.3.0 (#932)
• Remove the dependecy on a specific torchvision and torchmetrics packages.
• Bump PyTorch Lightning version to v.1.9.* (#870)
• Make input image normalization and center cropping configurable from config (#822)
• Improve flexibility and configurability of subset splitting (#822)
• Switch to new datamodules design (#822)
• Make normalization and center cropping configurable through config (#795)
• Switch to new changelog format. (#777)
• Rename feature to task (#769)
• make device configurable in OpenVINO inference (#755)
• 🚨 Fix torchmetrics version (#754)
• Improve NNCF initilization (#740)
• Migrate markdownlint + issue templates (#738)
• 🐞 Patch Timm Feature Extractor (#714)
• Padim arguments improvements (#664)
• 📊 Update DFM results (#674)
• Optimize anomaly score calculation for PatchCore (#633)

Deprecated

• Deprecated PreProcessor class (#795)
• Deprecate OptimalF1 metric in favor of AnomalyScoreThreshold and F1Score (#796)

Fixed

• Fix bug in anomalib/data/utils/image.py to check if the path is directory (#919)
• Fix bug in MVTec dataset download (#842)
• Add early stopping to CS-Flow model (#817)
• Fix remote container by removing version pinning in Docker files (#797)
• Fix PatchCore performance deterioration by reverting changes to Average Pooling layer (#791)
• Fix zero seed (#766)
• Fix #699 (#700)
• 🐞 Fix folder dataset for classification tasks (#708)
• Update torchmetrics to fix compute_on_cpu issue (#711)
• Correct folder mask path (#660)
• Fix >100% confidence issue for OpenVINO inference (#667)
• Update pre-commit links and some other minor fixes (#672)
• Fix black formatting issues. (#674)

New Contributors

• @jpcbertoldo made their first contribution in #660
• @tanemaki made their first contribution in #766
• @Jonas1302 made their first contribution in #771
• @yunchu made their first contribution in #885
• @strelka145 made their first contribution in #919

Full Changelog: v0.3.7...v0.4.0

Dobot Dataset and API

This is to release the assets for the Dobot The Magician, an education robot arm portable and capable to run various automation tasks. The release contains an API to control the robot and a sample dataset captured via the robot.

v0.4.0rc2

[v0.4.0rc2] - 2023-02-10

Added

• Add ShanghaiTech Campus video anomaly detection dataset (#869)
• Add pyupgrade to pre-commit configs, and refactor based on pyupgrade and refurb (#845)
• Add CFA model implementation (#783)
• Add RKDE model implementation (#821)
• Add Visual Anomaly (VisA) dataset adapter (#824)
• Add Synthetic anomalous dataset for validation and testing (#822)
• Add Detection task type support (#822)
• Add UCSDped and Avenue dataset implementation (#822)
• Add base classes for video dataset and video datamodule (#822)
• Add base classes for image dataset and image dataModule (#822)
• ✨ Add CSFlow model (#657)
• Log loss for existing trainable models (#804)
• Add section for community project (#768)
• ✨ Add torchfx feature extractor (#675)
• Add tiling notebook (#712)
• Add posargs to tox to enable testing a single file (#695)
• Add option to load metrics with kwargs (#688)
• 🐞 Add device flag to TorchInferencer (#601)

Changed

• Remove the dependecy on a specific torchvision and torchmetrics packages.
• Bump PyTorch Lightning version to v.1.9.* (#870)
• Make input image normalization and center cropping configurable from config (#822)
• Improve flexibility and configurability of subset splitting (#822)
• Switch to new datamodules design (#822)
• Make normalization and center cropping configurable through config (#795)
• Switch to new changelog format. (#777)
• Rename feature to task (#769)
• make device configurable in OpenVINO inference (#755)
• 🚨 Fix torchmetrics version (#754)
• Improve NNCF initilization (#740)
• Migrate markdownlint + issue templates (#738)
• 🐞 Patch Timm Feature Extractor (#714)
• Padim arguments improvements (#664)
• 📊 Update DFM results (#674)
• Optimize anomaly score calculation for PatchCore (#633)

Deprecated

• Deprecated PreProcessor class (#795)
• Deprecate OptimalF1 metric in favor of AnomalyScoreThreshold and F1Score (#796)

Fixed

• Fix bug in MVTec dataset download (#842)
• Add early stopping to CS-Flow model (#817)
• Fix remote container by removing version pinning in Docker files (#797)
• Fix PatchCore performance deterioration by reverting changes to Average Pooling layer (#791)
• Fix zero seed (#766)
• Fix #699 (#700)
• 🐞 Fix folder dataset for classification tasks (#708)
• Update torchmetrics to fix compute_on_cpu issue (#711)
• Correct folder mask path (#660)
• Fix >100% confidence issue for OpenVINO inference (#667)
• Update pre-commit links and some other minor fixes (#672)
• Fix black formatting issues. (#674)

Full Changelog: v0.4.0rc1...v0.4.0rc2