Embedded Vision Summit 2017: Thoughts From the Floor

kristof-667x500

This blog post was originally published at Imagination Technologies' website. It is reprinted here with the permission of Imagination Technologies.

Having recently returned from another exciting Embedded Vision Summit show we’d like to share our thoughts in a blog post. This year’s event once again impressed, with compelling conversations and discussions about vision and surrounding technologies setting the event apart from other less focused conferences.

After the show we came away with the feeling that machine learning was surely at the peak of the hype curve – and according to this graph below, Gartner seems to agree with us.

One of the key uses for machine learning is vision processing and there was clearly much interest in this subject from a very engaged crowd. The decision to have the vision technology showcase going on until 8.00 PM was a brave choice, but fears that it would be very quiet after 6.00 PM were quashed with crowds through to the very end – no doubt helped by the free beer and food provided!

This time round, we have taken the slightly different approach of obtaining the thoughts of various members of the Imagination staff who attended, gaining their impressions of the event via two simple questions.


Chris Longstaff, Senior Director of Marketing, PowerVR

What was the most interesting presentation you attended and why?

I enjoyed the Khronos OpenVX workshop because it showed the weight of the industry movement behind this open framework for Vision processing and the demand from people for an open standards-based solution for real-world deployable vision applications.

What was the most interesting demo you saw and why?

The Ai.Motive real-time simulator was an interesting demo because it showed the level to which graphics/CGI has reached and provided an alternative view of how the industry could go about training neural networks.


Francisco Socal, Technology Marketing Manager

What was the most interesting presentation you attended and why?

The coming shift from image sensors to image sensing”, by LG’s Paul Gallagher was great as it described very well the technology adoption phases for image sensors and computer/embedded vision compared to the adoption of previous technologies such as communications and general computing. It also showed ups where we are now and what to expect next.

What was the most interesting demo you saw and why?

Intel’s RealSense demos were interesting and seemed to have evolved over the last few years.


Steve Harris, VP Sales

What was the most interesting presentation you attended and why?

I was fascinated to learn how Val Marchevsky of Motorola explained how they use deep learning for ISP/lens/sensor tuning. They explained how it was possible for them to identify what needed to be tuned after it had been trained much better than humans could have done. They also use it for HDR decisions and quoted a 30% improvement on their traditional algorithms, something for which I hadn’t considered deep learning could be used.

What was the most interesting demo you saw and why?

I know this is a little biased but I really liked Imagination’s OpenVX CNN extension demo. Running on a Chromebook equipped with a Mediatek’s 8173 SoC featuring a PowerVR Series6XT GPU, it’s based on OpenVX 1.1 with CNN extensions and shows how the eco-system has evolved to make CNNs easily deployable for real-world vision applications. Click here for a full blog post giving more details.


Kristof Beets, Senior Director Business Development

What was the most interesting presentation you attended and why?

Although possibly a biased opinion, I really enjoyed Paul Brasnett of Imagination talking about the optimising training for CNNs to enable efficient inference. This is a key topic of interest in the industry and an area where much work is needed.

What was the most interesting demo you saw and why?

I was impressed with Qualcomm’s always-on vision solution to provide context awareness to devices, as it takes efficiency to the extreme. It is ultra-low-power and fully integrated with the sensor with just enough precision and processing to wake up the larger elements in the SoC or even the whole device based on the surroundings of the phone.


Mike Holland, Director of Sales

What was the most interesting presentation you attended and why?

I was fascinated with Rudy Berger, from Woodside Capital with his VC discussion, where he gave an insight into the commercial side of the vision industry.

What was the most interesting demo you saw and why?

From the Imagination demos, I was very interested to see how flexible the GPU is at running neural networks and that the PowerVR GPU achieves a good performance level. It was interesting to see how the differing networks produced different performance levels, both in terms of object recognition rate and object recognition accuracy. Read our blog post on the demo to see this in action.


Saraj Mudigonda, Director of Segment Marketing

What was the most interesting presentation you attended and why?

At Google’s TensorFlow presentation, Pete Warden highlighted two things: “Eight bits are enough” for CNNs, and how they are working closely with chip builders on TensorFlow and embedded platforms.

Another interesting presentation was from CheckVideo on intelligent video surveillance. The key points to note were first, that of the 300 million video security cameras in operation worldwide, less than five percent currently use vision capability. Secondly, that since many enterprises would like to use their legacy existing coax infrastructure there will be a resurgence of analogue HD video over coax, slowing down adoption of vision applications on-camera.

What was the most interesting demo you saw and why?

For me, it was the Uncanny Vision demo on the Renesas stand running their CV algorithms. They were showing smart fridges with face recognition technology. This indicated that a lot of these demos are still prototyped on the CPU: we foresee many of these being ported to run on faster processors, such as PowerVR GPUs, before real-world deployment.


Thanks to all my colleagues for this input. I look forward to the next Embedded Alliance member meeting to catch up with several new acquaintances met during this Summit.

Were you at the Embedded Vision Summit event?  How would you answer these two questions?

By Chris Longstaff
Senior Director of Product & Technology Marketing for PowerVR, Imagination Technologies

Here you’ll find a wealth of practical technical insights and expert advice to help you bring AI and visual intelligence into your products without flying blind.

Contact

Address

Berkeley Design Technology, Inc.
PO Box #4446
Walnut Creek, CA 94596

Phone
Phone: +1 (925) 954-1411
Scroll to Top