Edge AI and Vision Insights Newsletter
Edge AI and Vision Alliance
Inspiring and empowering engineers to design products that perceive and understand.
?LETTER FROM THE EDITOR
Dear Colleague,?
Next Thursday, October 17, 2024 at 9 am PT, SiMa.ai will present the free webinar “Delivering High Performance, Low Power Complete Edge AI Applications with the SiMa.ai One Platform MLSoC and Toolset” in partnership with the Edge AI and Vision Alliance. SiMa.ai has redefined AI at the edge. With software designed to ingest and accelerate any ML model in any modality from any framework, SiMa.ai MLSoC silicon provides a purpose-built platform for edge AI by enabling complete edge-ML application-pipeline acceleration on a single chip. Delivering data-center-class 50 TOPS performance with leading-edge power efficiency, and targeting applications such as industrial, retail, and healthcare, SiMa’s MLSoC accelerates deep learning workloads through a silicon-plus-software approach, streamlining the creation and deployment of intelligent edge AI solutions.
This webinar, presented by Carlos Davila, Director of Software Product Management, and Vidhyananth Venkatasamy, Principal Solutions Architect, details the unique architecture of the SiMa.ai MLSoC, highlighting how it overcomes traditional computing limitations, and describes the capabilities of the associated toolset:
Multiple live demonstrations will illustrate in detail the concepts presented in the webinar, and a question-and-answer session will follow the presentation. For more information and to register, please see the event page.
Brian Dipert
Editor-In-Chief, Edge AI and Vision Alliance
?
LEVERAGING INDUSTRY STANDARDS
For decades, the OpenCV software library has been popular for developing computer vision applications. However, developers have found it challenging to create efficient implementations of their OpenCV applications on processors optimized for edge applications, like the Qualcomm Snapdragon family. As part of its comprehensive support for computer vision application developers, Qualcomm provides a variety of tools to enable developers to take full advantage of the heterogeneous computing resources in the Snapdragon processors. In this talk, Xin Zhong, Computer Vision Product Manager at Qualcomm Technologies, introduces a new element of Qualcomm’s computer vision tools suite: a version of OpenCV optimized for Snapdragon platforms, which allows developers to leverage and port their existing OpenCV-based applications seamlessly to Snapdragon platforms. Supporting OpenCV v4.x and later releases, this implementation contains unique Qualcomm-specific accelerations of OpenCV and OpenCV extension APIs. Zhong explains how this library enables developers to leverage existing OpenCV code to achieve superior performance and power savings on Snapdragon platforms.
Offloading processing to accelerators enables embedded vision systems to process workloads that exceed the capabilities of CPUs. However, parallel processors add complexity as workloads must be distributed and synchronized across processing resources. Proven open standard APIs and languages can significantly streamline the design and programming of parallel processing systems, avoiding the need for companies to independently create and support an entire ecosystem of processors, tooling, developer education and supporting third-party libraries and machine learning stacks. In addition, effective open standards separate hardware and software development, enabling systems to integrate new components with ease, for cross-generation reusability and field upgradability. In this presentation, Neil Trevett, President of the Khronos Group, gives an update on Khronos Group open standards for programming and deploying accelerated inferencing, embedded vision and safety-critical systems, including the in-development Kamaros open standard API for embedded camera systems.
DEEP LEARNING TRAINING TECHNIQUES
In this presentation, Amit Mate, Founder and CEO of GMAC Intelligence, introduces the essential aspects of training convolutional neural networks (CNNs). He discusses the prerequisites for training, including models, data and training frameworks, with an emphasis on the characteristics of data needed for effective training. He explores the model training process using visuals to explain the error surface and gradient-based learning techniques. Mate’s discussion covers key hyperparameters, loss functions and how to monitor the health of the training process. He also addresses the common training problems of overfitting and underfitting, and offers practical rules of thumb for mitigating these issues. Finally, he introduces popular training frameworks and provides resources for further learning.
Developers of machine-learning based computer vision applications often face difficulties obtaining sufficient data for training and evaluating models. In this talk, Brian Geisel, CEO of Geisel Software, explores the use of synthetic data techniques to overcome these challenges. He explains the “simulation-to-reality” gap and the challenges it poses for realistic synthetic data generation. Using an automated Mars exploration vehicle application, Geisel shares how his company generates synthetic data using the Unity simulation environment, and demonstrates how synthetic data can be effective—while also highlighting its limitations. He also touches on the potential for few-shot learning techniques to reduce the need for training data.
UPCOMING INDUSTRY EVENTS
Delivering High Performance, Low Power Complete Edge-AI Applications with the SiMa.ai One Platform MLSoC and Toolset – SiMa.ai Webinar: October 17, 2024, 9:00 am PT
Your Next Computer Vision Model Might be an LLM: Generative AI and the Move From Large Language Models to Vision Language Models – Edge AI and Vision Alliance Online Symposium: October 23, 2024, 9:00 am PT
Embedded Vision Summit: May 20-22, 2025, Santa Clara, California
FEATURED NEWS
Intel Unveils Next-gen AI Solutions with the Launch of Xeon 6 and Gaudi 3
BrainChip Introduces Its Lowest-power AI Acceleration Co-processor
AMD Announces the Energy-efficient EPYC Embedded 8004 Series for Embedded Systems
Renesas Drives ADAS Innovation with its Power-efficient 4th-generation R-Car Automotive SoCs
The MIPI Alliance Releases A-PHY v2.0, Doubling the Maximum Data Rate of the Automotive SerDes Interface to Enable Emerging Vehicle Architectures
EDGE AI AND VISION PRODUCT OF THE YEAR WINNER SHOWCASE
Ambarella’s central 4D imaging radar architecture is the 2024 Edge AI and Vision Product of the Year Award Winner in the Edge AI Software and Algorithms category. It is the first centralized 4D imaging radar architecture that allows both central processing of raw radar data and deep low-level fusion with other sensor inputs—including cameras, lidar and ultrasonics. The central 4D imaging radar architecture combines Ambarella’s highly efficient 5 nm CV3-AD AI central domain controller system-on-chip (SoC) and the company’s Oculii adaptive AI radar software. This architecture’s optimized hardware and software provides the industry’s best AI processing performance per watt, for the lowest possible energy consumption, along with the most accurate and comprehensive AI modeling of a vehicle or robot’s surroundings. Ambarella’s Oculii AI radar algorithms uniquely adapt radar waveforms to the environment, achieving high angular resolution (0.5 degrees), an ultra-dense point cloud (10s of thousands of points/frame), and a long 500+ meters detection range, while using an order-of-magnitude fewer antennas for reduced data bandwidth and power consumption versus competing 4D imaging radars. Likewise, this architecture enables processor-less edge radar heads, further reducing both upfront costs and post-accident expenses (most radar modules are located behind the vehicle’s bumpers).
Please see here for more information on Ambarella’s central 4D imaging radar architecture. The Edge AI and Vision Product of the Year Awards celebrate the innovation of the industry’s leading companies that are developing and enabling the next generation of edge AI and computer vision products. Winning a Product of the Year award recognizes a company’s leadership in edge AI and computer vision as evaluated by independent industry experts.