Data acquisition for machine learning can be costly due to the large quantities of high-quality, unbiased data required to train a state-of-the-art model. Machine learning based object detection algorithms can require thousands of hand-labelled example images before they can reliably detect objects in images; collecting and labelling this data is a significant time investment.

This case study summarises the research completed by the University of Sheffield Advanced Manufacturing Research Centre (AMRC) Design and Prototyping Group’s digital design team to create a tool enabling the automated generation of labelled synthetic data for training machine learning object detection algorithms.