April 1, 2014 was the date for the Israel Machine Vision Conference (IMVC) in Tel Aviv. I’m always slightly wary of attending events held on April 1. I never know for sure that after all the queuing, checking in, travelling, waiting for bags, finding the hotel, the venue, someone doesn’t just say “April Fool!” when you get there. Well, not to worry… IMVC was very real. It’s the annual get-together for the prolific Israeli computer vision community and packed in a day of fascinating talks about the latest developments in this exciting subject. It was great to see so much innovation in evidence and to hear from companies large and small working in this area.
As with many areas of technology there was much talk about mobile. Computer vision in mobile devices is a hot topic, particularly as the energy efficient yet powerful processors required are quickly coming of age. Roberto Mijat and I were there to talk about ARM’s central role in this area and in particular the advantages of using the GPU to offload some of the processing required to enable these sorts of features.
In full flow talking about GPU Compute on Mali
Devices containing the ARM® Mali™-T600 series of GPUs have been providing general purpose compute capabilities for a couple of years now and there are many examples of the benefits of using the GPU for both graphics and non-graphics computation. I showcased a few of these in my talk, including GPU-accelerated gesture recognition from Israeli company eyeSight® Technologies and face detection and analysis from PUX (Panasonic), both of which have been optimised to run on the Mali-T604 GPU using OpenCL™. In these and many other cases we see the GPU making sufficient difference to enable computer vision algorithms to run in real time. Better still the GPU gives us the additional compute bandwidth which allows the use of more sophisticated algorithms that have been shown to enhance the user experience significantly. eyeSight’s low-light gesture detection is a great example. And equally as important is that we can do all this whilst burning much less energy – a crucial requirement for mobile devices.
eyeSight's gesture recognition in action (as shown at CES 2014)
Another area of discussion – both in my talk and elsewhere at the conference – compared the different ways of achieving computer vision on mobile. As well as using GPUs, ARM’s CPU processor technology already offers heterogeneous features through big.LITTLE™ and NEON™ technology, and there are custom DSPs designed for specific image processing jobs that can sit alongside the ARM CPU. A DSP is hard to beat when it comes to area and power, but the downside is its lack of flexibility. As new algorithms come along you need new DSPs – and this is where the programmable GPU really scores. It allows existing hardware to take on powerful new capabilities.
Demonstrating PUX face detection and analysis demo at IMVC
We met with many interesting companies and discussed some compelling new computer vision use cases boding well for what we’ll see emerging over the next few months. The conference ended with some interesting stargazing from both Google and Microsoft. And then it was all over for another year. Our hosts and conference organiser was SagivTech Ltd, a company dedicated to computer vision research, development and education. Our thanks to them for inviting us along, and for organising such a great event.
For more information, please visit: