From Harry Potter to Robust Regression: A career in computer vision and AI for visual effects, and a return to basic research

Abstract:
The first part of my talk covers some of my work in computer vision visual effects startups. My first company was formed by myself and a UniS student after my time as a UniS researcher. Working in the spare room of my house, we created the "Mocha" software. 25 years later, Mocha has won major awards and has been used on thousands of movies and TV shows. This and my other startup companies have addressed different commercial applications of computer vision, combining creative content with 3D and generative AI.

The second part discusses my current research in the area of robust model fitting. It may be somewhat surprising that there is active research in basic statistical problems such as mean estimation and regression analysis. When outliers are present in the data it is difficult to design algorithms that combine high statistical efficiency with good outlier rejection. I present some work in this area, an improvement to IRLS (Iteratively Reweighted Least Squares). I will then discuss possible directions and applications of this work.

Bio:
After a computer vision research career in the 1990s at various universities and finally UniS, working with Adrian Hilton and Phil Palmer, I switched to startup life with the idea to combine computer vision algorithms in an engineering context. My first startup, Imagineer Systems, was created with a student I met up with in my year as a UniS lecturer. The post-production video processing products we created won an Academy Award in 2013 and a Technology Emmy in 2019, both for the "planar tracker" algorithm I created in 1999. The "Mocha" software is part of Adobe Creative Suite and is used worldwide in TV and movie post-production. After Imagineer I worked at three startups as Head of R&D, creating computer vision and graphics technology that combine original creative content with 3D and AI elements. In 2026 I have decided to return to my roots as a researcher into low-level algorithms.