Bob Fisher

Bob Fisher is the first American recipient of the BMVA distinguished fellowship, but many in the UK computer vision community think of him as Scottish, as he has been at the University of Edinburgh for almost thirty years. Bob began his career as an undergraduate Mathematician at Caltech, moving to Stanford for his master’s. His academic career did not begin immediately, because he worked as a programmer for five years before coming to Edinburgh in 1983 to begin a PhD under Jim Howe. Edinburgh in the eighties was a world power in AI research, and vision was clearly established as one of the big AI challenges. At that time, the influence of Marr and Grimson meant that although everyone knew the “big problem” of object recognition was the ultimate goal, much attention was focussed on the low-level steps of recovering Marr’s “2½D sketch”, or other 3D representations (Grimson’s “From images to surfaces”). Bob looked beyond that goal, to a future where 3D would be available using real-time high resolution depth cameras, and his PhD thesis, entitled “From surfaces to objects”, explored the recognition of complex articulated 3D objects from range images. In common with most vision work of that era, the concepts in the thesis were proven on a relatively small dataset, but in this case there is a good reason why. Because these range cameras did not yet exist, Bob had to come up with an alternative: manual measurement using a tape measure of the distance between points in the scene and a 2D camera, followed by a custom surface interpolation technique, which made image collection rather more arduous than it is today: simply surfing Flickr. What he could not have known was how long it might take for technology to catch up. As we can see with this year’s release of a new computer game controller based on realtime depth sensing, Bob was about 30 years ahead of his time.

Those early papers were the first of some 250 publications in many areas of computer vision, including 3D sensing and calibration, surface modelling, least squares fitting, attention, evolutionary algorithms, and recently to medical imaging and video. His recent work displays an intriguing focus on animals and animal noises: humpback whalesong analysis and imaging of bats. Those wondering where the attraction lies in imaging bats may find some explanation in the fact that data collection for such studies is best achieved in exotic locations such as Panama.

I know Bob well, having studied for my PhD under his supervision, and recall his often ironically stated advice to PhD students. I’m sure that when he wrote “always ignore everything your supervisor says”, he realised the irony was evident, as to obey the instruction would be to fail to ignore a supervisor’s advice. In fact, one knew that it was an excellent idea not to ignore Bob’s advice. One of the many things I learned from him is a principle I hold dear to this day: ensure your paper is clear and easy to read—if an idea is worth publishing it’s worth explaining clearly and concisely with the minimum of extra fuss. I also learned that if one aims to solve the big or impossible problems, one will end up meeting and solving interesting problems along the way. I also learned that for any program we would write in Matlab or Perl, Bob could write the same program in less time in C! The value Bob places on clarity and in the promulgation of ideas is also visible in the vast amount of work he has done to make available computer vision teaching materials. Most of us will have used CVonline or HIPR at some point; and in the days before wikipedia it was a lifeline for researchers not just in the UK but throughout the world. His more recent work on the Dictionary of Computer Vision and on the CAVIAR video dataset is further testament to his commitment to the vision community.

It is therefore my very great pleasure to announce that Robert Fisher is the recipient of the 2010 Distinguished Fellowship of the BMVA.