Brandon Gilles's guest appearances

Brandon Gilles

The Amp Hour Electronics Podcast

The Engineering Entrepreneur Podcast

Across Kentucky

Bill Johnson

Scott Smith

Jan Witowski

Tony Whatley

Avidan Ross

Mitch McConnell

James Wilson

Anthony Wall

Sarah Brown

John Warrillow

Sam Stranks

Christina Cyr

Daymond John

John Davis

Jeff Perry

OwlTail

Daviess County's Brandon Gilles was a national finalist in the AFBF Discussion Meet.

Daviess County's Brandon Gilles was a national finalist in the AFBF Discussion Meet.

Daviess County's Brandon Gilles was a national finalist in the AFBF Discussion Meet

Brian Weinstein

Welcome Brandon Gilles and Brian Weinstein of Luxonis Holding Corporation  They make the DepthAI system, which uses the Intel Movidius Myriad X It was initially created for life safety problems, specifically targeted at distracted drivers. There is research that a car horn on a bike can help to lower fatal interactions. Brandon had multiple people in his life hurt or killed by cars while biking. Though a systemic change is needed, a tech solution is more likely in the short term. They built a hardware protoype and spent a couple hours of coding to prove out the concept. “Depth perception + CV is a cheat code”, but it needs to be useful in the physical world No way to embed this much power Current depth cameras on the market Intel D435 Intel D455   Two camera solution senses the disparity in the images AI (OpenCV) helps tell what is there (Car, bus, etc) Intel compute sticks Combining depth plus AI There are other Myriad X SOMs Jay Carlson episode where he was talking about SOMs and MIPI 2 lane vs 4 lane MIPI Base boards open source on github Luxonis did a Kickstarter with OpenCV and raised $1.4M Working with vendors Niche that didn’t exist: embedded, low cost, fast boot, performant OpenCV is a compiled library PyTorch DNN Module DARPA grand challenge OpenCV is a huge community The DepthAI has bindings for ROS/Python What is the Myriad X chip background? Intel bought the Movidius team in 2016 (Vox) 28 processors on the die, with a “Network on chip architecture“ What gets access to the data from the camera first? It is flexible, so the software configures it Asymmetric fell out of favor Geometries of the boards for the Kickstarter 0.4mm pitch, 400 pin BGA OAK-1 has the color camera, OAK-D has the color camera plus stereo vision December delivery, which they are on time for Possible to get the boards right now at higher prices They are working with Arducam to try and source different camera modules. Lower cost overall will come from lower cost cameras Camera module industry is very opaque This is why there is a custom RPi camera IMX 378 Apple and Android are doing visual depth preception on phones, including Apple’s new LiDAR on the iPhone Luxonis did a competition around spatial AI and got over 300 entries Many of the entries involved “giving another sense” to the visually impared. Following the bald referee Privacy centric What is edge computing? “Embedded is barely a thing yet for AI” OpenMV by Kwabena Agyeman works on a ST part They are working on building out the team. If you’re a PCB layout engineer proficient in Altium, send them an email. (remember: The Amp Hour always recommends sharing a portfolio of your work) SOM Info, including the pinout of the connector on the SOM, is on the github For more information, check out Luxonis.com  Many thanks to our Patrons! You can join at Patreon.com/TheAmpHour if you’d like to join the crowd. A special thanks to our corporate sponsor Binho, who now distribute the Sensepeek PCBite.

Welcome <a href="https://www.linkedin.com/in/brandon-gilles-5400b734/" rel="nofollow" target="_blank">Brandon Gilles</a> and <a href="https://www.linkedin.com/in/briandweinstein/" rel="nofollow" target="_blank">Brian Weinstein</a> of Luxonis Holding Corporation <ul> <li>They make <a href="https://shop.luxonis.com/collections/all/products/depthai-embedded-development-kit" rel="nofollow" target="_blank">the DepthAI system</a>, which uses the <a href="https://www.intel.com/content/www/us/en/products/processors/movidius-vpu/movidius-myriad-x.html" rel="nofollow" target="_blank">Intel Movidius Myriad X</a></li> <li>It was initially created for life safety problems, specifically targeted at distracted drivers.</li> <li>There is research that a car horn on a bike can help to lower fatal interactions.</li> <li><a href="https://www.crowdsupply.com/luxonis" rel="nofollow" target="_blank">Brandon had multiple people in his life hurt or killed by cars while biking.</a></li> <li>Though a systemic change is needed, a tech solution is more likely in the short term.</li> <li>They built a hardware protoype and spent a couple hours of coding to prove out the concept.</li> <li>“Depth perception + CV is a cheat code”, but it needs to be useful in the physical world</li> <li>No way to embed this much power</li> <li>Current depth cameras on the market <ul> <li><a href="https://www.intelrealsense.com/depth-camera-d435/" rel="nofollow" target="_blank">Intel D435</a></li> <li><a href="https://www.intelrealsense.com/depth-camera-d455/" rel="nofollow" target="_blank">Intel D455</a></li> </ul> </li> <li>Two camera solution senses the disparity in the images</li> <li>AI (<a href="https://opencv.org/" rel="nofollow" target="_blank">OpenCV</a>) helps tell what is there (Car, bus, etc)</li> <li><a href="https://www.intel.com/content/www/us/en/products/boards-kits/compute-stick.html" rel="nofollow" target="_blank">Intel compute sticks</a></li> <li>Combining depth plus AI</li> <li>There are other <a href="https://shop.luxonis.com/collections/all/products/bw1099" rel="nofollow" target="_blank">Myriad X SOMs</a></li> <li><a href="https://theamphour.com/515-embedded-linux-with-jay-carlson/" rel="nofollow" target="_blank">Jay Carlson episode where he was talking about SOMs</a> and MIPI</li> <li>2 lane vs 4 lane <a href="https://en.wikipedia.org/wiki/MIPI_Alliance" rel="nofollow" target="_blank">MIPI</a></li> <li><a href="https://github.com/luxonis/depthai" rel="nofollow" target="_blank">Base boards open source on github</a></li> <li><a href="https://www.kickstarter.com/projects/opencv/opencv-ai-kit" rel="nofollow" target="_blank">Luxonis did a Kickstarter with OpenCV and raised $1.4M</a></li> <li>Working with vendors</li> <li>Niche that didn’t exist: embedded, low cost, fast boot, performant</li> <li><a href="https://en.wikipedia.org/wiki/OpenCV" rel="nofollow" target="_blank">OpenCV is a compiled library</a></li> <li><a href="https://pytorch.org/" rel="nofollow" target="_blank">PyTorch</a></li> <li><a href="https://docs.opencv.org/master/d2/d58/tutorial_table_of_content_dnn.html" rel="nofollow" target="_blank">DNN Module</a></li> <li><a href="https://www.darpa.mil/about-us/timeline/-grand-challenge-for-autonomous-vehicles" rel="nofollow" target="_blank">DARPA grand challenge</a></li> <li>OpenCV is a huge community</li> <li>The DepthAI has bindings for <a href="https://www.ros.org/" rel="nofollow" target="_blank">ROS</a>/Python</li> <li>What is the Myriad X chip background?</li> <li><a href="https://www.intel.com/content/www/us/en/products/processors/movidius-vpu.html" rel="nofollow" target="_blank">Intel bought the Movidius team in 2016</a> (<a href="https://www.vox.com/2016/9/6/12810246/intel-buying-movidius" rel="nofollow" target="_blank">Vox</a>)</li> <li>28 processors on the die, with a “<a href="https://en.wikipedia.org/wiki/Network_on_a_chip#:~:text=The%20network%20on%20chip%20is,bus%20and%20crossbar%20communication%20architectures." rel="nofollow" target="_blank">Network on chip architecture</a>“</li> <li>What gets access to the data from the camera first?</li> <li>It is flexible, so the software configures it</li> <li>Asymmetric fell out of favor</li> <li>Geometries of the boards for the Kickstarter</li> <li>0.4mm pitch, 400 pin BGA</li> <li><a href="https://www.kickstarter.com/projects/opencv/opencv-ai-kit" rel="nofollow" target="_blank">OAK-1 has the color camera, OAK-D has the color camera plus stereo vision</a></li> <li>December delivery, which they are on time for</li> <li>Possible to get the boards right now at higher prices</li> <li>They are working with <a href="https://www.arducam.com/" rel="nofollow" target="_blank">Arducam</a> to try and source different camera modules.</li> <li>Lower cost overall will come from lower cost cameras</li> <li>Camera module industry is very opaque</li> <li>This is why there is <a href="https://www.amazon.com/Raspberry-Pi-Camera-Module-Megapixel/dp/B01ER2SKFS#:~:text=From%20the%20manufacturer,featuring%20a%20fixed%20focus%20lens.&amp;text=In%20terms%20of%20still%20images,1080p30%2C%20720p60%20and%20640x480p90%20video" rel="nofollow" target="_blank">a custom RPi camera</a></li> <li><a href="https://www.xda-developers.com/sony-imx378-comprehensive-breakdown-of-the-google-pixels-sensor-and-its-features/" rel="nofollow" target="_blank">IMX 378</a></li> <li>Apple and Android are doing visual depth preception on phones, including <a href="https://www.cnet.com/how-to/lidar-explained-iphone-12-pro-what-it-can-do-now-why-it-matters-for-the-future-ar-3d-scanning/" rel="nofollow" target="_blank">Apple’s new LiDAR on the iPhone</a></li> <li><a href="https://opencv.org/opencv-spatial-ai-competition/" rel="nofollow" target="_blank">Luxonis did a competition around spatial AI</a> and got over 300 entries</li> <li>Many of the entries involved “giving another sense” to the visually impared.</li> <li><a href="https://petapixel.com/2020/11/02/ai-tracking-camera-mistakes-referees-bald-head-for-a-soccer-ball/" rel="nofollow" target="_blank">Following the bald referee</a></li> <li>Privacy centric</li> <li>What is <a href="https://en.wikipedia.org/wiki/Edge_computing" rel="nofollow" target="_blank">edge computing</a>?</li> <li>“Embedded is barely a thing yet for AI”</li> <li><a href="https://openmv.io/pages/about" rel="nofollow" target="_blank">OpenMV by Kwabena Agyeman</a> works on a ST part</li> <li>They are working on building out the team.</li> <li>If you’re a PCB layout engineer proficient in Altium, send them an email. (remember: The Amp Hour always recommends sharing a portfolio of your work)</li> <li><a href="https://github.com/luxonis/depthai-hardware" rel="nofollow" target="_blank">SOM Info, including the pinout of the connector on the SOM, is on the github</a></li> <li>For more information, check out <a href="https://luxonis.com/" rel="nofollow" target="_blank">Luxonis.com</a></li> </ul> Many thanks to our Patrons! You can join at <a href="https://www.patreon.com/TheAmpHour" rel="nofollow" target="_blank">Patreon.com/TheAmpHour</a> if you’d like to join the crowd. A special thanks to our corporate sponsor <a href="https://binho.io/pages/sensepeek-accessories" rel="nofollow" target="_blank">Binho, who now distribute the Sensepeek PCBite</a>.

#517 – Depth and AI with Brandon Gilles and Brian Weinstein

Machine learning, Artificial intelligence and the OAK Project - Brandon Gilles -  Ep 89  Brandon is the founder and CEO of Luxonis and Chief Architect of the OpenCV AI Kit (OAK).  He's driven to build things that materially impact the world - and views OAK as his biggest opportunity to do so."  The OAK kit is a small chip with a camera that can be used to build your own AI machines. Say you want to remove a bad apple from a large pile? With OAK, you will be able to automate a robot to control this process.  We discuss a lot about how AI and machine learning work. If you are new to these topics, this will fill you in a lot.  We then discussed the specifics of Brandon’s background and how he created OAK.    His Kickstarter campaign just finished up and we discussed how that went as well.  Contact info:  Discussion Forum: https://discuss.luxonis.com/  Slack Community Invite: https://join.slack.com/t/luxonis-community/shared_invite/zt-emg3tas3-C_Q38TI5nZbKUazZdxwvXw  Email: brandon@luxonis.com 

Machine learning, Artificial intelligence and the OAK Project - Brandon Gilles -  Ep 89  Brandon is the founder and CEO of Luxonis and Chief Architect of the OpenCV AI Kit (OAK).  He's driven to build things that materially impact the world - and views OAK as his biggest opportunity to do so."  The OAK kit is a small chip with a camera that can be used to build your own AI machines. Say you want to remove a bad apple from a large pile? With OAK, you will be able to automate a robot to control this process.  We discuss a lot about how AI and machine learning work. If you are new to these topics, this will fill you in a lot.  We then discussed the specifics of Brandon’s background and how he created OAK.   His Kickstarter campaign just finished up and we discussed how that went as well.  Contact info:  Discussion Forum: <a href="https://discuss.luxonis.com/" rel="noopener nofollow" target="_blank">https://discuss.luxonis.com/</a> Slack Community Invite: <a href="https://join.slack.com/t/luxonis-community/shared_invite/zt-emg3tas3-C_Q38TI5nZbKUazZdxwvXw" rel="noopener nofollow" target="_blank">https://join.slack.com/t/luxonis-community/shared_invite/zt-emg3tas3-C_Q38TI5nZbKUazZdxwvXw</a> Email: <a href="mailto:brandon@luxonis.com" rel="noopener">brandon@luxonis.com</a>

Machine learning, Artificial intelligence and the OAK Project - Brandon Gilles -  Ep 89 