Developing a Machine Learning System for Reality Capture (R-CON Reality Capture Conference 2024)

Thomas Czerniawski, PhD is an applied machine learning scientist at Integrated Projects. In his presentation to the Reality Capture Network Conference in October of 2024, he covers the strategic, technical, and operational aspects of launching a Machine Learning (ML) initiatives in the reality capture industry.

ML provides an opportunity for massive operational efficiencies. Yet, the journey from conceptualization to successful implementation is fraught with challenges and questions. How do we initiate an ML program? Which processes can be automated and what are the prerequisites for leveraging ML effectively?

Drawing on our firsthand experience of developing a groundbreaking ML program for automating Scan-to-BIM processes, we offer a case study that illustrates not only the transformative potential of ML but also the practical steps and considerations involved in its application. We will explore essential topics such as the ideal timing and positioning for considering ML, the criteria for automation, necessary hardware, dataset requirements, and the critical decision of developing in-house versus purchasing solutions. Furthermore, we will discuss the human element—what expertise is needed, how to evaluate progress, manage risks, and decide when to pivot or persevere.

Our talk aims to equip attendees with the knowledge to navigate the ML landscape confidently, make informed decisions, and harness the power of ML to catalyze innovation within their organizations. By sharing our challenges and successes, we provide a candid look at the realities of integrating ML into reality capture, offering valuable insights for companies at any stage of their ML journey.