The Industrial Biscuits (Cookie) dataset is our internal dataset designed for the anomaly detection task, which captures Tarallini biscuits. It contains 1225 samples in four classes with the following structure:

• No defect (474 captions)
• Defect: not complete (465 captions)
• Defect: strange object (158 captions)
• Defect: color defect (128 captions)

In order to augmentate the dataset we rotated each sample by 90° for three times, so the augmented dataset contains 4900 captures in total. Augmented images were cropped to the size of their bounding boxes, the source samples remained uncropped in the original dataset.

EDIT This repository now contains refactored version (Version 2) of this dataset. The changed features are:

• All images are cropped and resized to the size 256x256
• Simplified annotations file without the bounding boxes (images are already cropped) and translated class names
• Added python script, which prepares the dataset for the anomaly detection task to the OK x NOK folder structure

Use the attached python script only with Version 2.

Please cite us in your following work as:

 @misc{karel horak_simon bilik_ondrej bostik_lukas kratochvila_tomas zemcik_2022, title={Industry Biscuit (Cookie) dataset}, url={https://www.kaggle.com/dsv/4311115}, DOI={10.34740/KAGGLE/DSV/4311115}, publisher={Kaggle}, author={Karel Horak and Simon Bilik and Ondrej Bostik and Lukas Kratochvila and Tomas Zemcik}, year={2022} }