Benchmark Dataset for Tape Fibre Layups

This dataset contains surface scans of fibre tape layups for benchmarking evaluation algorithms and training Machine Learning algorithms that detect defects in these scans. It consists of a set of 1200 real scans, collected using a laser line triangulation sensor, and 1200 artificial images that aim to mimic the real scans. All images have pixel-accurate annotations.

For both subsets, a split into training data (990 images) and test data (210 images) has been predefined. For benchmarking purposes, it is recommended only to use the test data split of the real scans. All data points can be used for training and evaluating machine learning algorithms.