🚀 Implement MVTecAD2 dataset

[samet-akcay]

📝 Description

This PR adds support for the MVTec AD 2 dataset, which is an advanced version of the original MVTec AD dataset designed for unsupervised anomaly detection. The implementation includes support for three different test sets:

• Public test set (test_public/): Contains both normal and anomalous samples with ground truth masks
• Private test set (test_private/): Contains unseen test samples without ground truth
• Private mixed test set (test_private_mixed/): Contains unseen test samples with mixed anomalies without ground truth

Features

• Full implementation of the MVTecAD2 dataset class with support for all test types
• Proper handling of dataset splits (train, validation, test)
• Support for image augmentations and transforms
• Comprehensive test coverage for all test types and edge cases
• Integration with existing Anomalib data pipeline

✨ Changes

• Added MVTecAD2 datamodule in src/anomalib/data/datamodules/image/mvtecad2.py
• Added MVTecAD2Dataset class in src/anomalib/data/datasets/image/mvtecad2.py
• Added TestType enum to handle different test set types
• Added unit tests for the dataset implementation
• Updated dummy dataset generation for testing
• Added configuration example in examples/configs/data/mvtecad2.yaml

Select what type of change your PR is:

• 🐞 Bug fix (non-breaking change which fixes an issue)
• 🔨 Refactor (non-breaking change which refactors the code base)
• 🚀 New feature (non-breaking change which adds functionality)
• 💥 Breaking change (fix or feature that would cause existing functionality to not work as expected)
• 📚 Documentation update
• 🔒 Security update

Testing

The implementation includes comprehensive tests that verify:

• Dataset loading and preprocessing
• All three test types (public, private, mixed)
• Image augmentations and transforms
• Error handling for invalid test types
• Dataset split functionality
• Integration with the Anomalib pipeline

Usage Example

from anomalib.data import MVTecAD2

# Create datamodule with public test set
datamodule = MVTecAD2(
    root="./datasets/MVTec_AD_2",
    category="sheet_metal",
    train_batch_size=32,
    eval_batch_size=32,
)

# Use private test set
datamodule = MVTecAD2(
    root="./datasets/MVTec_AD_2",
    category="sheet_metal",
    test_type="private",
)

# Access different test sets
datamodule.setup()
public_loader = datamodule.test_dataloader()  # returns loader based on test_type
private_loader = datamodule.test_dataloader(test_type="private")
mixed_loader = datamodule.test_dataloader(test_type="private_mixed")

License

The MVTec AD 2 dataset will be released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License (CC BY-NC-SA 4.0).

Notes

• This implementation requires proper dataset access and follows the MVTec AD 2 dataset license terms
• Users need to download the dataset separately from the MVTec website
• The implementation maintains compatibility with the existing Anomalib API

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB