A new thread that I will fill up. Why? Its because of yesterdays webinar. A LOT more of the IoT than I had even considered is/will be image recognition. THere was a foreshadowing phrase from BF in there somewhere.... QUIK will be returning to image work- as a very general paraphrase. Nepes is doing a fair bit of work in this area...their authentication phrases were very interesting to me... General paraphrase. If you combine facial recognition with voice ID its a really good authentication 100% Nepes- we can do the facial recognition part, but don't have voice in house. So I will put things of a general nature around the pixel processing and the IoT to learn and try to get a glimpse of where we might be going.... I want to put up the definition of the adjacent possible again soon- the idea is that you just go from one room to the next and open a door in that one. You don't want to do a Babbage and go up 50 floors ahead of everybody else. I will have a few comments on quikAI and will sprinkle them in as I post items that relate. jfieb
building up a foundation Image Recognition and the Internet of Everything Physical objects that are recognized and identified through computer-aided vision are magically bound to the digital world. January 4, 2015 by Bill French Chief Architect at Intelliscape The field of computer-aided vision and image recognition took a significant step forward in 2012. And since then, it has become apparent that computer-aided vision technology has established a permanent cornerstone in the future of software applications. How this relates to the Internet of Things (IoT) is so obvious that few have given it a second thought. And even fewer have considered the emerging prospects of the Internet of Recognition™ - an idea based on pervasive machine vision. Our basic understanding of IoT predicates that devices of many types will soon have IP addresses and the capacity to communicate with other devices and applications over networks. Indeed, it is our fundamental understanding that members of the IoT must possess embedded electronics to participate as bona-fide “things” on the network. This is a rational if not a narrow definition of IoT. Reliable, automated image recognition is about to change that definition. Common Objective IoT and image recognition share a common goal. IoT's primary purpose is to create a bond between the physical world and digital space. Image recognition events create virtual representations of actual physical objects - a virtualization of "things". The subtle relationship between objects that can be recognized without human assistance and IoT is as elegant as it is invisible. ______________________________ Physical objects that are recognized and identified through computer-aided vision are magically bound to the digital world. ______________________________ In some cases, physical objects of an image recognition event are homogeneous - i.e., a Tide laundry logo on a race car, a bus advert, or perhaps the logo on an actual box of Tide detergent. But it's also possible that certain recognition events may contain uniquely identifiable artifacts. For example, a medical device whose serial number is legible in the image, or the image of a vehicle with a license tag. Image recognition events fall into these two basic categories with the former a prime candidate for aggregate analytics, and the latter serving as the basis for discrete identity, i.e., a unique member in the emerging internet of things. Homogenous Recognition Events This type of recognition event possesses properties that are likely to be useful in aggregate contexts. Indeed, the recognition of a class of automobile parts which have failed, establishes a place in the IoT not as a specific “thing”, but rather a “class” of things. Attributes of the class - such as the same automobile brake part failing repeatedly - are able to convey understanding and awareness at an aggregate level. Intelligent software need only monitor the class to know when to recommend a safety recall. This is no different than an IP-enabled heat pump informing a service center that a motor is in need of maintenance. The IoT must embrace classes of things as readily as discrete things. Failure to do so rules out the benefits that can be derived from a large and compelling aspect of big data. Discrete Recognition Events Image recognition events that identify a specific thing instantly transform physical objects into unique members of the IoT. Such events are precise and able to capture time, place, and condition of a specific physical object. These objects are immutable or at least near-immutable in the context they exist. Camera technology has become so inexpensive and connectivity so pervasive, that it's economically practical to assign cameras to specific physical objects to monitor performance, location, and condition. Combined with recognition software, the camera itself becomes an IoT proxy for the physical object and is thus capable of detecting change.this is what Nepes is doing now and what they showed us yesterday. ______________________________ Image recognition technology has the capacity to swiftly embrace physical objects that were never designed to participate in the Internet of Things.______________________________ Read-Only Objects? Despite the ability to include inert physical artifacts in the IoT, recognized objects lack the capacity to listen for and act upon digital instructions. Certainly, computer-aided vision technology is good, but it can't magically will a motor to repair itself. In fact, most IP-enabled devices will require human intervention for many situations and for decades to come. However, the proxy relationship established through recognition software has the capacity to compel other systems and people to take action upon these inert objects. The cost of repeated monitoring through image detection is extremely low, giving us the economic latitude to perform millions of observation events. Indeed, recognition events create read-only instantiations within the IoT, but awareness about these events can be used to indirectly effectuate change to the objects. The Internet of Everything There's a massive wave of data coming to the IoT from images and video. Perhaps today's big data movement is prologue to a future when the Internet of Things is transformed into the Internet of Everything. For this transformation to occur it will largely depend on image recognition technology that enables identity by visual proxy.What did I learn?Somehow I really like the idea that an effort is never wasted even if it was not a success. QUIK knows it can process Pixelsreally well, they will be back at it again, but with a better ending?Also this is a whole lot of fun.
THey have this very well mapped out...what is done on each Eos core. FFE-accelertor FPGA i will focus a little there as I want to glimpse what it can do... feature extraction. Key thing look at the bigger heading...Neuromem and FPGA together....
I kind of enjoyed the Nepes segment. I think it was the real world examples that helped bring the message across in a way I can understand. The Eos will be with these guys as they inspect the 22FDX-SOI as they roll off the lines at Fab 1 in Dresden?
here is Nepes authentication work...it was somewhere in here that Nepes spoke of the need for working together with someone as they have the image stuff, they do NOT have the voice and they need it. What have I learned already.... 1. QUIK will be doing Pixel work again, and that good as its a BIG part of the IoT. No I will not rush it, but it is in the cards. 2. The Eos S3 can do it! On that one I am so happy, I had a fear that it would take the S4. 3. FPGA is doing VITAL stuff, not just important chores 4. It fits a key thing of S Johnsons work...these things are here and now; they are within arms reach or a short stroll down the hallway of a room they are already in. This is NOT Facebook's thought reading research that is being done...that is 50 floor up, nothing here is Babbage.
General commentary. So much work done in the year. It is impressive, thnks for the hard work at QUIK everryone- the most since I have owned a part of this company. eFPGA has a dedicated tools team in Bangalore and has invested time to make sure these tools are ready and easy to use. A lot of trips to Asia for those important proof of concept hearables that made CES of 18 the best one they have been to. this new initiative that BF has put together is comprehensive, it fits together very well with no overlap. You can get a glimpse that like the eFPGA software they want it to be easy to use to get stuff monitored. It is AI, and while the eyes of the market are only looking at the cloud, at beating the worlds best go player, at Nividia, tensor flow etc. I will use this thread to try to grasp the amazing range and sheer size of AI on the far edge. The eFPGA is doing VITAL stuff and not just some menial connectivity tasks. That is sticky, its margin, nice to see real world appreciation for eFPGA at the edge. It helps to have a story to tell that resonates with folks. AI on the edge, with just a few of the right slots can be that story. Again for me I am very happy. I have been waiting for this. QUIK has been at work, since that job opening of 2 yrs ago- they said NOTHING, just did the work- same as when they did the initial work on the FFE in the S1, said nothing and did the work, someone else might have had a similar idea- I hope BF has others to put their shoulders to the wheel to do some of these things... who holds the hand of the tier 1 and delivers on the changes they ask for again and again, and works to convert the other devices into slots. who goes to China over and over to work with the important customers that want a familiar face? who goes to India to see the very very important eFPGA tools team? who puts a crucial AI IoT platform together? goes to investor conferences? how much can one person do? This thread will be FUN for me, but it will be a TON of new material and for those who are busy with work and life, just skip it. If I find something really crucial in some way I will bring it to the forums attention on the busy thread of the time. To Brian Faith and all the QUIK folks A thank you again for such hard work. We don't see that so often.
