‘Deep Learning and Vision Algorithm Development in MATLAB Targeting Embedded GPUs’

The EMVA proudly announces that Mr. Alexander Schreiber, Principal Application Engineer at MathWorks, confirmed to share his expertise on Deep Learning at Embedded VISION Europe conference.

Alexander Schreiber joined the Application Engineering department of MathWorks Germany in 2008 and works as Principal Application Engineer and Technical Account Manager. He covers application areas of autonomous systems design, HW/SW co-design, automatic code generation and verification (HDL, C). Prior to joining MathWorks he worked as ASIC Library Designer, ASIC Designer and Project Manager for EDA and semiconductor companies. He holds a M.Sc. equivalent degree in Electrical Engineering from the University of Stuttgart.

Abstract of  Alexander’s presentation:

Learn how to adopt a MATLAB centric workflow to design, verify and deploy your computer vision and deep learning applications on to embedded Tegra-based platforms including Jetson TK1/TX1 and DrivePX boards. The workflow starts with algorithm design in MATLAB, which enjoys universal appeal among engineers and scientists because of its expressive power and ease-of-use. The algorithm may employ deep learning networks augmented with traditional computer vision techniques and can be tested and verified within MATLAB. Next, a compiler auto-generates portable and optimized CUDA code from the MATLAB algorithm, which is then cross-compiled and deployed to the Tegra board. The workflow affords on-board real-time prototyping and verification controlled through MATLAB. Examples of common computer vision algorithms and deep learning networks are used to describe this workflow, and their performance benchmarks are presented.

The debut of EMVA’s brandnew conference Embedded VISION Europe, supplemented by an already well booked table top exhibition, will take place 12-13 October 2017 in Stuttgart.

Find all conference details at www.embedded-vision-emva.org

‘New image processing paradigma in embedded vision:  Camera-integrated 3D laser triangulation’

The EMVA proudly announces that Dr. Thomas Däubler, CTO at NET New Electronic Technology, will present a use case at the first edition of Embedded VISION Europe.

Thomas Däubler joined NET New Electronic Technology GmbH as CTO in 2015. He shapes NET’s roadmap incl. industrial and medical cameras under consideration of new technologies and market demand.  In future, NET’s open camera concept will provide a wider basis for camera-embedded vision solutions. Prior to joining NET, Thomas worked in product development, product management, and business development for test systems and machine vision solutions. He holds a PhD in physics from University of Potsdam.

Abstract of  Thomas’ presentation:

Smart vision camera manufacturer NET New Electronic Technology has supported digMAR, an Austrian solution provider, to develop a custom embedded 3D inspection system for textile cuttings. Due to the extra-wide web and fast inspection speed conventional image processing solutions (PC-controlled camera systems) are not capable to process the big data streams in time on the PC. The open camera concept allows a considerable reduction of data to be processed by the PC for real-time quality control. This is done by shifting tasks from PC to the camera-embedded FPGA of NET´s standard GigE Vision cameras. The open camera concept enables customers to implement own algorithms in the camera or smart vision system for unique embedded vision solutions without supplementary vision hardware.

The debut of EMVA’s brandnew conference Embedded VISION Europe, supplemented by an already well booked table top exhibition, will take place 12-13 October 2017 in Stuttgart.

Find all conference details at www.embedded-vision-emva.org

‘reVISION Accelerating Embedded Vision and Machine Learning applications at the Edge’

The EMVA could win Mr. Giles Peckham, Regional Marketing Director at XILINX, to speak at the debut edition of Embedded VISION Europe.

Giles has more than 30 years’ experience in the semiconductor industry, starting with the design of ASSPs for consumer applications at NXP before moving on to FAE and marketing roles for gate array and standard cell products and finally a sales role in the same organisation.  After five years in international marketing and sales roles at European Silicon Structures, the e-beam direct-write ASIC company, Giles recognised the increasing potential for FPGAs and joined Xilinx.
Whilst at Xilinx, he has held a number of technical and commercial marketing roles in EMEA and now runs a global marketing team from his office in London. Giles holds a BSC in Electronic Engineering and Physics from Loughborough University, UK and a Professional Postgraduate Diploma in Marketing from the Chartered Institute of Marketing in the UK.

Abstract of Giles’ presentation:

The traditional approach to developing programmable logic based embedded vision and machine learning systems is to first use a high-level modelling language such as Open VX / OpenCV or Caffe to define the algorithm. Once the algorithm has been defined, it is then recreated by a specialist team within a HDL, targeted for the selected programmable logic device. This approach introduces a disconnection between the high-level algorithm and the implemented algorithm which significantly increases development time, programme risk, and NRE cost. What is needed is the ability to work with high-level, industry standard frameworks and libraries without the need to rewrite the algorithm into a specific HDL at a lower level.
The reVISION™ acceleration stack which supports both All Programmable Zynq® UltraScale+™ MPSoC and Zynq®-7000 SoC developments, addresses these challenges and eliminates the gap. reVISION provides support for both OpenVX and OpenCV in the embedded vision sphere and Caffe for machine learning. At the core of the reVISION stack is the SDSoC™ tool which enables system level development of the Zynq-7000 and Zynq MPSoC using high-level languages such as C, C++ and OpenCL™. SDSoC, as a system optimising compiler, enables the designer to identify bottlenecks which impact performance within the processing system once the algorithm has been developed, and accelerate these into the programmable logic. This acceleration is performed without the need for an HDL specialist. This is made possible thanks to SDSoC’s combination of Vivado® High-Level Synthesis and a connectivity framework to seamlessly move functions between the processing system and the programmable logic. To support this, reVISION provides several acceleration capable OpenCV functions (including the OpenVX Core Subset) for embedded vision and machine learning inference engine elements.
Embedded vision developers are therefore able to leverage the benefits of using these SoC devices, and the image processing pipeline can be implemented within the device’s programmable logic. Using the acceleration capable OpenCV functions enables the development of the algorithm once in an industry standard framework. This frees up the processing system to be used to implement the higher level, decision making algorithms and system / communication functions. When it comes to decision making, reVISION provides support for Caffe and can take a prototxt file to define a Convolutional Neural Network, which can be implemented within the programmable logic.
Such an approach to both embedded vision and machine learning removes system bottlenecks and produces a system which is more responsive, power efficient and reconfigurable than a traditional GPU/CPU based approach.

The debut of EMVA’s brandnew conference Embedded VISION Europe, supplemented by an already well booked table top exhibition, will take place 12-13 October 2017 in Stuttgart.

Find all conference details at www.embedded-vision-emva.org

Our newest online educational initiative regarding the 1288 standard for camera characterization will be composed by two elements.

As a first step short video clips will be broadcasted weekly within the next four weeks, each clip explaining one highlight of the new release 3.1 of the standard.
Please find the video clips under www.emva.org/news-media/media/videos/ .

Following a free one hour interactive webinar is offered on Tuesday, September 12, 2017:
“How to compare cameras with the new EMVA 1288 standard, Release 3.1“.
Participants may ask questions during the webinar and afterwards by e-mail.
Please Register Now

In addition to the online initiative a new series of training courses given in English and German language are prepared in cooperation with two of our EMVA members. Find the details at:
www.aeon.de/en/emva-1288.html
and
www.framos.com/en/the-emva-1288-camera-standard-new-version-3.1

The EMVA proudly announces that Mr. Marco Jacobs, Vice President Marketing at videantis GmbH, will give his presentation titled ‘Demystifying embedded vision processing architectures’ at Embedded VISION Europe.

Marco Jacobs, VP of marketing at videantis, has over 20 years of experience in the semiconductor IP industry and video/imaging applications. At videantis, he is responsible for corporate and product marketing and works with key semiconductor manufacturers to bring novel, higher-quality computer vision and video applications to its customers.

Prior to joining videantis, Marco was VP of marketing at Vector Fabrics, director of multimedia marketing at ARC, held management positions at semiconductor startups Silicon Hive (acquired by Intel) and BOPS, and was a software architect at Philips. Marco studied computer science at the Delft University of Technology in the Netherlands and at the University of North Carolina, Chapel Hill. He holds 7 issued patents.

Abstract of Marco’s presentation:

Computer vision algorithms have made tremendous progress and are rapidly becoming crucial components in new applications such as automotive ADAS and self-driving cars, mobile phones, AR/VR, IoT, smart surveillance cameras, and drones. Even simple computer vision algorithms require huge performance and to enable these to run on embedded, small and low power devices requires specialized architectures.
In this talk we will analyze typical computer vision algorithms and show how they’re different from typical workloads. We will then give an overview of the different processing architectures that can be used to implement computer vision algorithms: CPUs, GPUs, vision processors, FPGAs, and hard-wired accelerators. We’ll show trade-offs and typical performance, power, and cost factors. Finally, we’ll look at how these components are integrated into system-level architectures.
.

The debut of EMVA’s brandnew conference Embedded VISION Europe, supplemented by an already well booked table top exhibition, will take place 12-13 October 2017 in Stuttgart.

Find all conference details at www.embedded-vision-emva.org

Platinum Conference Sponsor

The EMVA proudly announces Mr. David Moloney, Director of machine vision technology, NTG, at Intel Corporation, talking about ‘Low-cost Edge-based Deep Learning Inference and Computer Vision in Consumer and Industrial Devices’ at Embedded VISION Europe.

David was formerly the chief technology officer of Movidius, which he founded in 2005 with Sean Mitchell.

He has a BEng in electronic engineering from Dublin City University and a PhD from Trinity College Dublin in the area of FPGA-based HPC for computational fluid dynamics. He has worked in the semiconductor industry internationally for the past 28 years with Infineon in Germany, STMicroelectronics in Italy, ParthusCeva (Ceva-DSP) and Frontier Silicon in Ireland.
Moloney has 31 granted patents and numerous publications. He acts as a reviewer for IEEE, a communications magazine, and for the EU Commission on programmes such as ARTEMIS. He has collaborated on many EU initiatives, including the Eyes of Things (EoT) Horizon 2020 project.
His interests include: processor architecture; computer vision algorithms; hardware acceleration and systems; hardware and multiprocessor design for DSP communications; and HPC and multimedia applications.

Abstract of David’s presentation:

There has been rapid progress in terms of network accuracy on datasets such as ImageNet since Alex Krizhevsky’s breakthrough AlexNet paper in 2012. The trend has been to ever deeper and more complex networks with greater computational and memory requirements as we creep up the asymptote. On a separate track many companies are anxious to get these networks out of the lab and into products which imposes some tension in terms of trading off accuracy against what can reasonably achieved in an embedded platform. Balancing this tension involves close cooperation between advanced R&D teams developing networks and engineering teams developing Vision Processing Units that can operate at the network edge.  We will describe how these tensions can be resolved and how advanced R&D into CNNs is influencing both the field of embedded computer vision and the hardware these vision systems run on.

The debut of EMVA’s brandnew conference Embedded VISION Europe, supplemented by an already well booked table top exhibition, will take place 12-13 October 2017 in Stuttgart.

Find all conference details at www.embedded-vision-emva.org

Platinum Conference Sponsor