Hardware Podcasts

221. Generational Discovery & The New Age of Computer Vision

The Next Byte Podcast · 3 Jun 2025 · 18m

AI in practice: one real use of AI described

What this episode covers

The hosts discuss research from Princeton University and the University of Washington, led by a professor and his PhD student, on using metasurfaces to revolutionize computer vision. They explain how these nanostructured, flat lenses pre-filter light to perform hardware-level edge detection, achieving a 99% reduction in computational power compared to traditional software-based methods. The technology, which can be as small as a grain of salt, beats industry benchmarks like AlexNet and enables faster, more efficient object detection for applications like autonomous vehicles and smart home devices.

AI in practice

Where this conversation describes AI being used to build or deliver a product. Part of the AI in practice tracker.

Design99.4% less computer power than usual

The team used machine learning to determine the optimal layout of nano-pillars for the metasurface lens, automating the design process to filter light effectively before it hits the sensor.

“the team figure out a way to use machine learning to help design the best layout of pillars. So they're not just manually designing every single part in this matrix manually by themselves. They're basically using machine learning to tell them what they need to do to filter the light…”

Speaker 2 · Hear it at 15:48

Most relevant to

early-stage companies

Topics

Mentioned

Companies
Mouser Electronics, Princeton University, University of Washington, Tesla, Toyota Research Institute, Stanford, DeepMind, Anthropic
Also
AlexNet

Related episodes from other shows

Other shows on the same ground, closest first.

  • Embedded.fm · 1h 24m

    477: One Thousand New Instructions

    Kwabena Agyeman, founder of OpenMV, explains how SIMD optimization and new Arm Cortex-M55 processors enable complex computer vision and machine learning on low-cost microcontrollers. He details specific algorithmic improvements, such as a 16x speedup in median filters, and discusses the shift from Linux-based systems like NVIDIA Jetson to tinyML for cost and power efficiency. Agyeman also shares his experience scaling OpenMV through the chip shortage and his previous career at Embark Trucks.

    Embedded firmwareMicrocontrollers & dev platforms
    8 chapters

  • Embedded.fm · 1h 11m

    161: Magenta Doesn't Exist

    Kat Scott, a senior software engineer at Planet and co-author of 'Practical Computer Vision with SimpleCV', explains the fundamentals of computer vision, emphasizing that it is largely about discarding irrelevant data to find specific signals. She details the challenges of industrial computer vision, noting that consistent lighting and constrained environments are critical for success, and warns that achieving high accuracy in machine learning requires exponentially more data for each decimal place of improvement. Scott also discusses the technical pipeline behind real-time face filters like those on Snapchat, which involve face detection, 3D mesh fitting, and reprojection, and advises engineers to use tools like OpenCV and ROS rather than reinventing the wheel.

    AI & machine learningIndustrial & consumer product design
    10 chapters

  • EE Times Current · 43m

    The Seer of Prophesee | CES: The Good, The Bad, and The Weird…, and The Sleepable, The Driveable, The Mixological

    Luca Verre, CEO of Prophesee, explains how their event-based sensors capture only changes in a scene to reduce data bandwidth for automotive and industrial applications, highlighting a collaboration with Sony on a new 1-megapixel sensor. EE Times editors then review CES 2020, discussing privacy chips from D-Clock, edge processing software from Tiraki, and autonomous boats from Brunswick. They also cover IBM's quantum computing farm, John Deere's AI-powered precision agriculture sprayers, and various consumer gadgets like smart beds and AI toothbrushes.

    Sensors & measurement in productsAI & machine learning
    8 chapters

  • Embedded.fm · 1h 24m

    497: Everyone Likes Tiny

    Kwabena Agyeman, President and Co-Founder of OpenMV, discusses the launch of two new programmable camera systems, the OpenMV Cam N6 and the OpenMV AE3, via a Kickstarter campaign. He explains how integrating Neural Processing Units (NPUs) and Arm Helium SIMD instructions enables high-performance computer vision and edge AI on microcontrollers, achieving significant speedups while maintaining ultra-low power consumption. Agyeman details the software ecosystem, including MicroPython, TensorFlow Lite for Microcontrollers, and partnerships with Edge Impulse and Roboflow, which simplify model training and deployment for embedded developers.

    Embedded firmwareMicrocontrollers & dev platforms
    8 chapters

  • EE Times Current · 46m

    AI Revolutionizes Video Capture | “The Current War” Reviewed | V2X Babel

    EE Times editors Brian Santo and George Leopold review the film 'The Current War,' discussing the rivalry between Thomas Edison and George Westinghouse over DC and AC power, and the pivotal role of Nikola Tesla. The episode then covers French startup Prophesee, which has developed neuromorphic event-based vision sensors and raised $28 million to mass-produce its chips for automotive and IoT applications. Finally, the hosts analyze the industry split between DSRC and 5G for vehicle-to-everything (V2X) communication, noting Volkswagen's adoption of DSRC versus the 5G backing of BMW and Ford.

    Chips, FPGAs & semiconductorsAI & machine learning
    5 chapters

The summary, chapters and topics are generated from the episode’s audio by this directory, not written by The Next Byte Podcast. The episode itself belongs to its publisher.