Image dataset for AI-based screw detection

The dataset enables the training of a neural network to recognise screw heads and missing screws (aka holes) in and on electronic devices for automated disassembly. It was developed as part of the Desire4Electronics project and is used to recognise screws so that they can be automatically loosened by an industrial robot.

The dataset comprises 821 images of autonomous vacuum cleaners. Its labels comprise screw heads (Torx, Allen and Phillips types) and holes and are in LabelMe format with bounding boxes.

For reference, please cite this dataset's DOI as well as the project publication this dataset is an unofficial supplement of. This dataset was gathered and labelled at the Fraunhofer Institute for Manufacturing Engineering and Automation IPA.