This repository provides the OLP-dataset underlying our manuscript Increasing the Generalization of Supervised Fabric Anomaly Detection Methods to Unseen Fabrics published at Sensors in 2022 as well as accompanying code for data-loading.
OLP consists of 38 patterned fabrics. Opposed to related, public fabric defect datasets, it consists of high-resolution image-pairs taken with both front-light and back-light illumination. Such a multi-illumination setup was shown to improve performance at detecting defects, and is state-of-the-art in industry. Furthermore, we provide instance-level defect classes, segmentations and bounding boxes.
For exhaustive details, we refer to the overview given in the manuscript.
The dataset can be downloaded and extracted by executing:
wget https://www.lfb.rwth-aachen.de/download/olp-dataset.tar
tar -xvf olp-dataset.tar -C olp-dataset/ && rm olp-dataset.tar
in a shell. Alternatively, the URL can of course be opened in the browser to trigger the download that way.
We provide the images of every fabric in the corresponding Texile_$ID folder, and store all annotations in Texile_$ID/dataset.json.
Annotations are furthermore given in the popular ms-coco format.
The design-philosophy is to then construct a dataset per fabric, and compose them via concatenation.
For examples, please look at the two notebooks provided in dataloading-code.
If you make use of the OLP-dataset, please consider citing our paper published at Sensors
@Article{s22134750,
AUTHOR = {Rippel, Oliver and Zwinge, Corinna and Merhof, Dorit},
TITLE = {Increasing the Generalization of Supervised Fabric Anomaly Detection Methods to Unseen Fabrics},
JOURNAL = {Sensors},
VOLUME = {22},
YEAR = {2022},
NUMBER = {13},
ARTICLE-NUMBER = {4750},
URL = {https://www.mdpi.com/1424-8220/22/13/4750},
ISSN = {1424-8220},
DOI = {10.3390/s22134750}
}
If you wish to contact us, you can do so at [email protected]
Copyright (C) 2022 by RWTH Aachen University
http://www.rwth-aachen.de
License:
This software is dual-licensed under:
• Commercial license (please contact: [email protected])
• AGPL (GNU Affero General Public License) open source license
The dataset itself is furthermore licensed under CC-BY-ND-SA-4.0.