Release Anomalib Library v2.6.0

What's Changed

Added

• 🚀 model: Add SuperADD VAND4.0 winning method by @alexriedel1 in #3628
• 🚀 data: Add AutoVI Dataset by @julie-bat in #3633

Removed

• ⚠️ data: Remove Kaputt deprecated APIs (use_reference, reference_only, category='all'); use image_mode=ImageMode.* and category=None / a concrete category (default book_other) instead
• ⚠️ metric: Remove deprecated AUPRO num_thresholds constructor argument and unused compute_pro(..., target=...) parameter
• ⚠️ utils: Remove temporary resolve_with_warning / _is_legacy_default_root helpers; use resolve_dataset_root instead

Changed

• 🔧 data: Kaputt datamodule default category is now book_other (was the deprecated all pseudo-category)

Fixed

• 🐞 data: Distribute split remainder over subsets to avoid ZeroDivisionError by @prawnsgupta in #3680
• 🐞 security: Fix path traversal by @AlexanderBarabanov in #3686
• 🔧 deps: Bump torch to resolve Dependabot alerts by @AlexanderBarabanov in #3681

New Contributors

• @prawnsgupta made their first contribution in #3680
• @julie-bat made their first contribution in #3633

Full Changelog: lib/v2.5.1...lib/v2.6.0

---

API Changes

New Public Exports

The following class is now exported from anomalib.models:

Class	Module Path
SuperADD	anomalib.models.image.super_add

from anomalib.data import MVTecAD2
from anomalib.engine import Engine
from anomalib.models import SuperADD

datamodule = MVTecAD2(category="sheet_metal")
model = SuperADD(patch_size=640, patch_overlap=128)
engine = Engine()
engine.fit(model=model, datamodule=datamodule)

The following classes are now exported from anomalib.data:

Class	Module Path
AutoVI	anomalib.data.datamodules.image.autovi
AutoVIDataset	anomalib.data.datasets.image.autovi

from anomalib.data import AutoVI

datamodule = AutoVI(
    root="./datasets/AutoVI",
    category="pipe_clip",
)
datamodule.prepare_data()  # downloads only the zip for `pipe_clip`
datamodule.setup()

Breaking Changes

These APIs were deprecated in prior releases and are removed in v2.6.0:

Removed API	Replacement	Since
Kaputt(category='all')	category=None (all categories) or a concrete material (default is now book_other)	v2.4.0
KaputtDataset(reference_only=...)	image_mode=ImageMode.REFERENCE_ONLY	v2.4.0
KaputtDataset(use_reference=...)	image_mode=ImageMode.QUERY_AND_REFERENCE	v2.4.0
AUPRO(num_thresholds=...)	Omit the argument (it was ignored)	v2.1.0
AUPRO.compute_pro(..., target=...)	compute_pro(cca=..., preds=...) only	v2.1.0
resolve_with_warning / _is_legacy_default_root	resolve_dataset_root	v2.4.x

Kaputt category and image mode

# Before (removed)
from anomalib.data import Kaputt

datamodule = Kaputt(category="all", use_reference=True)
datamodule = Kaputt(reference_only=True)

# After
from anomalib.data import Kaputt
from anomalib.data.datasets.image.kaputt import ImageMode

datamodule = Kaputt(category=None)  # all categories
datamodule = Kaputt()  # default category is now "book_other"
datamodule = Kaputt(image_mode=ImageMode.QUERY_AND_REFERENCE)
datamodule = Kaputt(image_mode=ImageMode.REFERENCE_ONLY)

# Before (removed)
from anomalib.data.datasets import KaputtDataset

dataset = KaputtDataset(use_reference=True)
dataset = KaputtDataset(reference_only=True)

# After
from anomalib.data.datasets import KaputtDataset
from anomalib.data.datasets.image.kaputt import ImageMode

dataset = KaputtDataset(image_mode=ImageMode.QUERY_AND_REFERENCE)
dataset = KaputtDataset(image_mode=ImageMode.REFERENCE_ONLY)

AUPRO

# Before (removed)
from anomalib.metrics import AUPRO

metric = AUPRO(fpr_limit=0.3, num_thresholds=100)
fpr, pro = metric.compute_pro(cca=cca, preds=preds, target=masks)

# After
from anomalib.metrics import AUPRO

metric = AUPRO(fpr_limit=0.3)
fpr, pro = metric.compute_pro(cca=cca, preds=preds)

Dataset root resolution

# Before (removed)
from anomalib.utils.path import resolve_with_warning

root = resolve_with_warning("./datasets/MVTecAD", "MVTecAD")

# After
from anomalib.utils.path import resolve_dataset_root

root = resolve_dataset_root("./datasets/MVTecAD", "MVTecAD")
root = resolve_dataset_root(None, "MVTecAD")  # -> ~/.cache/anomalib/datasets/MVTecAD

Existing Deprecations (carried forward)

These were deprecated in prior releases and remain deprecated:

Deprecated API	Replacement	Since	Removal Target
anomalib.models.components.dinov2	TimmFeatureExtractor	v2.5.1	Future
Legacy ONNX exporter (dynamo=False)	Dynamo exporter (dynamo=True)	v2.5.1	v2.7.0
Engine.export(ov_args=...)	Use ExportParameters directly	v2.3.0	Future
PatchcoreTorchModel.compute_distance(embeddings=...)	Default memory bank path	v2.1.0	Future

# Prefer (current)
from anomalib.models.components.feature_extractors import TimmFeatureExtractor

extractor = TimmFeatureExtractor(
    backbone="vit_small_patch14_dinov2",
    layers=["blocks.2", "blocks.5", "blocks.8", "blocks.11"],
)

# Avoid (deprecated)
from anomalib.models.components import dinov2  # FutureWarning

# Prefer dynamo ONNX export
model.to_onnx(export_root="./exports", dynamo=True)

# Avoid legacy exporter (removed in v2.7.0)
model.to_onnx(export_root="./exports", dynamo=False)

What's Changed

• chore(lib): bump version by @ashwinvaidya17 in #3677
• chore(deps): update python:3.13-slim docker digest to 6771159 by @oep-renovate[bot] in #3678
• fix(deps): bump torch to resolve Dependabot alerts (v2) by @AlexanderBarabanov in #3681
• chore(deps): update dependency js-yaml to v4.3.0 [security] by @oep-renovate[bot] in #3682
• 🐞 fix(data): distribute split remainder over subsets to avoid ZeroDivisionError by @prawnsgupta in #3680
• ci: Add cleanup old images workflow by @ivanzati in #3683
• chore(deps): update dependency dompurify to v3.4.12 [security] by @oep-renovate[bot] in #3685
• chore(ci): add temp guard to prevent some Dependabot alerts from reopening by @AlexanderBarabanov in #3684
• fix: path traversal by @AlexanderBarabanov in #3686
• feat(lib): merge 2.6.0 branch by @ashwinvaidya17 in #3694

New Contributors

• @prawnsgupta made their first contribution in #3680

Full Changelog: lib/v2.5.1...lib/v2.6.0