TD-Magic: From Pictures of Timing Diagrams To Formal Specifications

Abstract

We introduce TD-Magic, the first neuro-symbolic approach for translating an image of a timing-diagram (TD) to a formal specification. We overcome the lack of labelled data for supervised learning, by first developing a synthetic data generator of labelled TDs. We then use object detection techniques to identify rising and failing edges, OCR to recognise the text, and image processing algorithms to capture synchronisation patterns. Finally, we use semantic interpretation to analyse the extracted features and generate the associated formal specification. Our experiments on industrial TDs show high translation accuracy opening the way to more sophisticated requirements-extraction algorithms from pictures.