Tabular datamodule (Custom dataset from DataFrame, CSV, or Parquet)

CHANGELOG.md

@@ -11,6 +11,7 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
 - 🚀 Add new SOTA video Anomaly detection module FUVAS
 - 🚀 Add VAD dataset by @abc-125 in https://github.com/open-edge-platform/anomalib/pull/2603
 - 🚀 Add Tiled Ensemble for V2 by @blaz-r in https://github.com/open-edge-platform/anomalib/pull/2660
+- 🚀 Add Tabular datamodule by @manuelkonrad in https://github.com/openvinotoolkit/anomalib/pull/2713
 
 ### Removed
 

docs/source/markdown/guides/reference/data/datamodules/image.md

@@ -28,6 +28,13 @@ Dataset format compatible with Intel Geti™.
 Custom folder-based dataset organization.
 :::
 
+:::{grid-item-card} Tabular
+:link: anomalib.data.datamodules.image.Tabular
+:link-type: doc
+
+Custom tabular dataset.
+:::
+
 :::{grid-item-card} Kolektor
 :link: anomalib.data.datamodules.image.Kolektor
 :link-type: doc

docs/source/markdown/guides/reference/data/datamodules/image/tabular.md

@@ -0,0 +1,7 @@
+# Tabular Datamodule
+
+```{eval-rst}
+.. automodule:: anomalib.data.datamodules.image.tabular
+   :members:
+   :show-inheritance:
+```

examples/configs/data/tabular.yaml

@@ -0,0 +1,73 @@
+class_path: anomalib.data.Tabular
+init_args:
+  name: bottle
+  root: "datasets/MVTecAD/bottle"
+  train_batch_size: 32
+  eval_batch_size: 32
+  num_workers: 8
+  test_split_mode: from_dir
+  test_split_ratio: 0.2
+  val_split_mode: same_as_test
+  val_split_ratio: 0.5
+  seed: null
+  samples:
+    - image_path: train/good/000.png
+      label_index: 0
+      mask_path: ""
+      split: train
+    - image_path: train/good/001.png
+      label_index: 0
+      mask_path: ""
+      split: train
+    - image_path: train/good/002.png
+      label_index: 0
+      mask_path: ""
+      split: train
+    - image_path: train/good/003.png
+      label_index: 0
+      mask_path: ""
+      split: train
+    - image_path: train/good/004.png
+      label_index: 0
+      mask_path: ""
+      split: train
+    - image_path: test/broken_large/000.png
+      label_index: 1
+      mask_path: ground_truth/broken_large/000_mask.png
+      split: test
+    - image_path: test/broken_large/002.png
+      label_index: 1
+      mask_path: ground_truth/broken_large/002_mask.png
+      split: test
+    - image_path: test/broken_large/004.png
+      label_index: 1
+      mask_path: ground_truth/broken_large/004_mask.png
+      split: test
+    - image_path: test/good/000.png
+      label_index: 0
+      mask_path: ""
+      split: test
+    - image_path: test/good/001.png
+      label_index: 0
+      mask_path: ""
+      split: test
+    - image_path: test/good/003.png
+      label_index: 0
+      mask_path: ""
+      split: test
+    - image_path: test/broken_large/001.png
+      label_index: 1
+      mask_path: ground_truth/broken_large/001_mask.png
+      split: test
+    - image_path: test/broken_large/003.png
+      label_index: 1
+      mask_path: ground_truth/broken_large/003_mask.png
+      split: test
+    - image_path: test/good/002.png
+      label_index: 0
+      mask_path: ""
+      split: test
+    - image_path: test/good/004.png
+      label_index: 0
+      mask_path: ""
+      split: test

src/anomalib/data/__init__.py

@@ -61,6 +61,7 @@
     MVTecAD2,
     MVTecLOCO,
     RealIAD,
+    Tabular,
     Visa,
 )
 from .datamodules.video import Avenue, ShanghaiTech, UCSDped, VideoDataFormat
@@ -75,6 +76,7 @@
     KolektorDataset,
     MVTecADDataset,
     MVTecLOCODataset,
+    TabularDataset,
     VADDataset,
     VisaDataset,
 )
@@ -181,6 +183,7 @@ def get_datamodule(config: DictConfig | ListConfig | dict) -> AnomalibDataModule
     "MVTecAD2",
     "MVTecLOCO",
     "RealIAD",
+    "Tabular",
     "VAD",
     "Visa",
     # Video Data Modules
@@ -196,6 +199,7 @@ def get_datamodule(config: DictConfig | ListConfig | dict) -> AnomalibDataModule
     "KolektorDataset",
     "MVTecADDataset",
     "MVTecLOCODataset",
+    "TabularDataset",
     "VADDataset",
     "VisaDataset",
     "AvenueDataset",

src/anomalib/data/datamodules/__init__.py

@@ -4,7 +4,7 @@
 # SPDX-License-Identifier: Apache-2.0
 
 from .depth import Folder3D, MVTec3D
-from .image import VAD, BTech, Datumaro, Folder, Kolektor, MVTec, MVTecAD, Visa
+from .image import VAD, BTech, Datumaro, Folder, Kolektor, MVTec, MVTecAD, Tabular, Visa
 from .video import Avenue, ShanghaiTech, UCSDped
 
 __all__ = [
@@ -16,6 +16,7 @@
     "Kolektor",
     "MVTec",  # Include MVTec for backward compatibility
     "MVTecAD",
+    "Tabular",
     "VAD",
     "Visa",
     "Avenue",

src/anomalib/data/datamodules/image/__init__.py

@@ -10,6 +10,7 @@
 - ``MVTecAD``: MVTec Anomaly Detection Dataset
 - ``MVTecAD2``: MVTec Anomaly Detection Dataset 2
 - ``MVTecLOCO``: MVTec LOCO Dataset with logical and structural anomalies
+- ``Tabular``: Custom tabular dataset with image paths and labels
 - ``VAD``: Valeo Anomaly Detection Dataset
 - ``Visa``: Visual Anomaly Dataset
 
@@ -36,6 +37,7 @@
 from .mvtecad import MVTec, MVTecAD
 from .mvtecad2 import MVTecAD2
 from .realiad import RealIAD
+from .tabular import Tabular
 from .vad import VAD
 from .visa import Visa
 
@@ -54,6 +56,7 @@ class ImageDataFormat(str, Enum):
     - ``MVTEC_AD_2``: MVTec AD 2 Dataset
     - ``MVTEC_3D``: MVTec 3D AD Dataset
     - ``MVTEC_LOCO``: MVTec LOCO Dataset
+    - ``TABULAR``: Custom Tabular Dataset
     - ``REALIAD``: Real-IAD Dataset
     - ``VAD``: Valeo Anomaly Detection Dataset
     - ``VISA``: Visual Anomaly Dataset
@@ -69,6 +72,7 @@ class ImageDataFormat(str, Enum):
     MVTEC_3D = "mvtec_3d"
     MVTEC_LOCO = "mvtec_loco"
     REAL_IAD = "realiad"
+    TABULAR = "tabular"
     VAD = "vad"
     VISA = "visa"
 
@@ -83,6 +87,7 @@ class ImageDataFormat(str, Enum):
     "MVTecAD2",
     "MVTecLOCO",
     "RealIAD",
+    "Tabular",
     "VAD",
     "Visa",
 ]