About
Philip O'Connor
Postgraduate Researcher
UCD
Philip O Connor is a PhD researcher focused in synthetic data and its applications in developing computer vision systems with artificial intelligence (AI). He graduated from University College Dublin (UCD) with both bachelor's and master's degrees in Mechanical Engineering, and later earned a master’s degree in Artificial Intelligence from the University of Galway.
Prior to beginning his PhD studies, he worked as an engineer at Intel for seven years, where he gained valuable experience in automation within manufacturing environments and the integration of vision systems into decision-making processes.
His research sits at the intersection of engineering and AI, with a strong industry focus. The project aims to advance the integration of AI-driven vision systems into a broader range of engineering applications and is co-sponsored by Digital Manufacturing Ireland (DMI).
Technical Summary
Current neural networks used in computer vision are typically trained on millions of images of everyday objects. While these networks can be adapted for specialised applications, such as production line monitoring, doing so requires large quantities of labelled images and significant human effort. The aim of this research is to leverage synthetic data—such as rendered images of 3D models—to generate training datasets, enabling the rapid deployment of new systems without the time-consuming and laborious process of collecting real-world images. This approach reduces the lag between production changes and the implementation of automated quality inspections.
In addition to synthetic images, other forms of data can also be incorporated, such as object distances, orientations, and spatial relationships. Since the data are generated computationally, the research offers full control over environmental factors like lighting, occlusions, camera positions, and lens characteristics. This allows for the creation of highly challenging scenarios during training, enabling neural networks to perform reliably under diverse and complex real-world conditions.
Privacy and security are further advantages of using synthetic data. As the images can be generated entirely from random, procedurally created objects, there are no privacy concerns related to data collection or usage. This approach also eliminates the need for human labelling and may even remove the requirement for long-term data storage, as datasets can be generated on demand. Such capabilities mark a significant shift from current large datasets, which often rely on a mix of human-labelled and automatically labelled data.