May 2016 Embedded Vision Summit Introductory Presentation (Day 2)

Copyright © 2016 Embedded Vision Alliance 1

Welcome to Day 2 of the Embedded Vision Summit!

Santa Clara, California May 3, 2016

Jeff Bier, Founder, Embedded Vision Alliance / President, BDTI


• Jeff Dean (Google):

• Deep learning has permeated all aspects of Google’s operators

• You can implement deep neural networks efficiently by training

with 32-bit floats and then quantize to 8-bit integers

• Bruce Daley (Tractica):

• Deep learning software market > $10B by 2024, driven

by apps like ad services, agriculture, manufacturing

• Chris Rowen (Cadence):

• Approximately 100% of sensor data will be image sensor data

• >100x energy and 20x performance from network & architecture optimization

What We Learned Yesterday


• Raj Talluri (Qualcomm):

• 8.7B smartphones will be shipped in next 5 years …

every single one of which is a computer vision platform

• Peter Shannon (Firelake):

• Computer vision is inductive reasoning problem …

and this makes vision software development fundamentally

different from traditional software development

• Paul Kruszewski (WRNCH):

• Video game industry blossomed once tools, middleware, and framework

matured … and the same will happen in computer vision



• Andreas Gal (Silk Labs):

• Increasingly powerful embedded processors give system developers the

option of putting visual intelligence at the edge instead of the cloud … and

this has important training and privacy considerations

• Stefan Heck (NAUTO):

• The actual cost of driving a car is $3/mile, 99% waste, and results in 33,000

fatalities a year … but computer vision can help

• Allen Rush (AMD):

• Computer vision will be essential to high quality augmented/virtual reality



• While we’re in the early days of vision software development infrastructure, vision

provides compelling benefits in a wide range of applications

• Deep learning was everywhere

• CNN tutorial spent a day on it (slides will be available)

• Vision Technology Showcase (you can’t swing a dead cat without putting it in

front of a CNN demo that identifies it as a dead cat… and if you don’t believe

me, just go there—it’s open from 10 am to 7 pm)

• Mentioned in almost every Business Insights track presentation yesterday



Using Neural Networks for Vision in Products?

Yes, extensively 15%

Yes, in a minor role 17%

Not yet, but planning to do

so 29%

No 29%

Don't know 10%

Source: Embedded Vision Alliance

Embedded Vision Developer Survey, Nov. 2015

N=417


Which Vision Libraries and APIs Do You Use?

63%

18%

6% 3%

11%

84%

47%

28%

21%

14%

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

OpenCV OpenGL OpenVX FastCV Other

1st Choice

Top 3

Source: Embedded

Vision Alliance

Embedded Vision Developer

Survey, Nov. 2015

N=385


• Auviz — AuvizVCA, video content analysis platform for FPGAs

• Auviz Systems & PLDA Group — FPGA-based

computer vision accelerators with QuickPlay

• Cadence —Tensilica Vision P6 DSP

• Khronos — Released OpenVX 1.1 specification

• Movidius — Fathom Neural Compute Stick and

Fathom deep learning software framework

• Qualcomm — Snapdragon Machine Learning SDK

Product Announcements at the Summit


Tuesday, May 3 — Day Two

Lunch

Welcome Remarks

Keynote: Using Vision to Enable Autonomous Land, Sea and Air Vehicles

1:00 PM

2:00 PM

1:00 PM

9:00 AM

10:45 AM

Vision Technology Showcase Reception 7:30 PM

3:00 PM

10:30 AM

Plenary Session: Computer Vision 2.0: Where We Are and Where We’re Going

Technical Insights Track

Business Insights Track

Enabling Technologies Track

Vision Technology Showcase (10:00 AM-7:30 PM)

Technical Insights Track

Business Insights Track

Enabling Technologies Track

Vision Technology Showcase 5:30 PM

5:30 PM


Wednesday, May 4 — Day Three

Lunch 12:00 PM

1:00 PM

9:00 AM

12:00 PM

Vision Technology Workshops

Synopsys Workshop:

Designing Low-power, Low-cost Vision Solutions Using

DesignWare EV Processors

Khronos Workshop:

Accelerate Your Vision Applications with OpenVX

Designing Low-power, Low-cost Vision Solutions Using

DesignWare EV Processors

Accelerate Your Vision Applications with OpenVX

1:00 PM

5:00 PM

Alternate OpenVX Workshop:

Tuesday, May 17, 2:00 - 6:00 PM AMD, Sunnyvale

bit.ly/23mbN9N


Floorplan


Thanks to Our Lead Sponsors


Embedded Vision Alliance Member Companies


Using Vision to Enable Autonomous

Land, Sea and Air Vehicles

Larry Matthies

Senior Scientist

NASA Jet Propulsion Laboratory

Tuesday Keynote