The MVTec 3D Anomaly Detection Dataset (MVTec 3D-AD)
MVTec 3D Anomaly Detection Dataset (MVTec 3D-AD) is a comprehensive 3D dataset for the task of unsupervised anomaly detection and localization. It contains over 4000 high-resolution scans acquired by an industrial 3D sensor. Each of the 10 different object categories comprises a set of defect-free training and validation samples and a test set of samples with various kinds of defects. Precise ground-truth annotations are provided for each anomalous test sample.
We provide a download link for the whole dataset, as well as links for each object category. For each object, the data consists of four folders:
'train', which contains the defect-free training samples.
'validation', which contains the defect-free validation samples.
'test', which contains the defective and defect-free test samples.
'calibration', which contains the internal camera parameters of the 3D sensor.
Each dataset split contains two subdirectories:
'xyz', which contains 3-channel TIFF images that store the x,y, and z coordinates.
'rgb', which contains 3-channel PNG images that store the corresponding RGB values for each 3D point.
The test split additionally includes a ground truth directory 'gt'. For each test sample, this directory contains a 1-channel PNG image. It indicates, for each image pixel, whether a defect is present or not. The defect type is reflected in the pixel value of the image. The mapping between defect names and pixel values is specified in the file class_ids.json.
In order to ensure a fair and consistent comparison of new and existing methods on our dataset, we provide python scripts that allow an easy evaluation. For details on how to use the script, please have a look at the included readme file.