MVTec AD

Format specification

The MVTec AD format specification is available here.

The dataset has annotations for detecting abnormal pixels through binary masks and it turns into bounding boxes or abnormal labels for supporting classification, detection, and segmentation tasks. The MVTec AD dataset is composed of training data only for good category without any annotation and testing data for both good and multiple defective categories with masks. The dataset contains total 15 kinds of objects or textures.

Supported tasks / formats:

• The combined format - mvtec

• Image classification - mvtec_classification

• Object detection - mvtec_detection

• Instance segmentation - mvtec_segmentation

Supported annotation types:

• Label (classification)

• Bbox (detection)

• Mask (segmentation)

Import MVTec AD dataset

The MVTec AD dataset is available for free download here.

A Datumaro project with a MVTec AD source can be created in the following way:

datum project create
datum project import --format mvtec_segmentation <path/to/dataset>

It is possible to specify project name and project directory. Run datum project create --help for more information.

The MVTec AD dataset directory should have the following structure:

└─ Dataset/Category
   ├── train/
   │   ├── good/ # directory with list of good images
   │   │   ├── img1.png
   │   |   ├── img2.png
   │   |   └── ...
   ├── test/
   │   ├── good/ # directory with list of good images
   │   │   ├── img1.png
   │   |   ├── img2.png
   │   |   └── ...
   │   ├── defective1/ # directory with list of defective images
   │   │   ├── img1.png
   │   |   ├── img2.png
   │   |   └── ...
   │   ├── defective2/ # directory with list of defective images
   │   │   ├── img1.png
   │   |   ├── img2.png
   │   |   └── ...
   └── ground_truth/ # directory with semantic segmentation masks
       ├── defective1/ # directory with list of defective images for detection and segmentation task
       │   ├── img1_mask.png
       |   ├── img2_mask.png
       |   └── ...
       ├── defective2/ # directory with list of defective images for detection and segmentation task
       │   ├── img1_mask.png
       |   ├── img2_mask.png
       |   └── ...

To make sure that the selected dataset has been added to the project, you can run datum project info, which will display the project information.

Export to other formats

Datumaro can convert a MVTec AD dataset into any other format Datumaro supports.

Such conversion will only be successful if the output format can represent the type of dataset you want to convert, e.g., image classification annotations can be saved in ImageNet format, but not as COCO keypoints.

There are several ways to convert a MVTec AD dataset to other dataset formats:

datum project create
datum project import -f mvtec <path/to/mvtec>
datum project export -f coco -o <output/dir>

or

datum convert -if mvtec -i <path/to/mvtec> -f coco -o <output/dir>

Or, using Python API:

import datumaro as dm

dataset = dm.Dataset.import_from('<path/to/mvtec>', 'mvtec')
dataset.export('save_dir', 'coco', save_media=True)

Export to MVTec AD format

There are several ways to convert an existing dataset to Pascal VOC format:

# export dataset into MVTec AD format (classification) from existing project
datum project export -p <path/to/project> -f mvtec -o <output/dir> -- --tasks classification

# converting to MVTec AD format from other format
datum convert -if imagenet -i <path/to/dataset> \
    -f mvtec -o <output/dir> \
    -- \
    --save-media

Extra options for exporting to MVTec AD format:

• --save-media - allow to export dataset with saving media files (by default False)

• --tasks TASKS - allow to specify tasks for export dataset, by default Datumaro uses all tasks.

datum project export -f mvtec -- --tasks detection,classification

Examples

Examples of using this format from the code can be found in the format tests.