Synthetic data generation can shorten the time for training deep learning models, but it can also require a lot of tweaking and testing to get synthetic data close to the real one. This session expands on the successful version 1.0 to bring together researchers working with digital twins, synthetic data from game engines and generative models for synthetic data generation. Our focus is on looking into use cases and best practices for synthetic images and videos, but also generating tabular data, annotations, 3D models, point clouds, and others.

We saw a lot of interest in synthetic data generation for medical use cases, for surface inspection, for damage detection, for surveillance, among others. There is also a lot of interest in creating synthetic data for very specific use cases, like vehicle and ship tracking, as well as the use of Nerfs and Gaussian Splatting for generation of novel synthetic views. This demonstrated that researchers and companies in Denmark and the Nordics are interested in the limitless possibilities coming from generating their own custom datasets and cutting the time from ideation to deployment.

Program

Session opening – 5 min talk, Kamal Nasrollahi, Director of Research

“Trustworthy AI with GraphRAG and Synthetic Embedding Generation” – 30 min presentation, Peter Damm, Director of Applied Intelligence at Milestone Systems

Chosen researchers that will present their research topic – 10 to 15 minutes –presentation depending on number.

Organizers

• Ivan Nikolov – Assistant Professor of Computer Graphics and Computer Vision, Aalborg University – iani@create.aau.dk
• Kamal Nasrollahi – Director of Research, Milestone Systems/ Professor of Computer Vision and Machine Learning, Aalborg University – kn@create.aau.dk

Level

Intermediate: for attendees who have basic understanding or some experience with the subject but are not yet advanced. We will be discussing research in various fields using synthetic and generative data. It will require understanding of machine learning but minimal prior knowledge for synthetic data and generative data.