I recently had the opportunity to present my latest research at the European Conference on Visual Media Production (CVMP) 2024, where researchers and industry professionals from the creative sectors come together to explore and discuss the latest advances in visual media production.

This year, I presented my paper, PDFed, which is a blockchain-based federated learning system for training diffusion models, addressing a critical privacy challenge – it enables us to measure and minimize the memorization of training data in generative AI models. I received thoughtful questions that reinforced the importance of this work to others in the field.

The other papers presented covered a wide range of topics, including generative modelling of human motion, image-based relighting, facial geometry reconstruction, and novel view synthesis.

It was also great to see my supervisor and DECaDE centre director, John Collomosse, present Adobe and our lab’s collaborative work for safe and fair generative AI, highlighting our research on content provenance and authenticity alongside the technical tools we are introducing to make these safeguards possible. I got to see how our individual projects come together into a cohesive body of work addressing content provenance and generative AI’s challenges from various angles, also considering the interplay between technology and law.

The conference itself was, as always, a good balance between industry-focused presentations from creative studios and academic research in visual production. The topics of the keynotes ranged from AI’s role in VFX to 3D digital human avatars.

It has been great to see how our work can be quickly integrated in creative studio pipelines for more efficient, inclusive and immersive experiences. Congratulations to the CVMP organizers for hosting another successful conference, and my thanks for the opportunity to present and engage with this community!