OPTIMIZING DEEP NEURAL NETWORK EFFICIENCY |
 |
|
DNN Quantization: Theory to Practice Deep neural networks, widely used in computer vision tasks, require substantial computation and memory resources, making it challenging to run these models on resource-constrained devices. Quantization involves modifying DNNs to use smaller data types (e.g., switching from 32-bit floating-point values to 8-bit integer values). Quantization is an effective way to reduce the computation and memory bandwidth requirements of these models, and their memory footprints, making it easier to run them on edge devices. However, quantization does degrade the accuracy of CNNs. In this talk, Dwith Chenna, Member of the Technical Staff and Product Engineer for AI Inference at AMD, surveys practical techniques for DNN quantization and shares best practices, tools and recipes to enable you to get the best results from quantization, including ways to minimize accuracy loss. |
 |
|
Leveraging Neural Architecture Search for Efficient Computer Vision on the Edge In most AI research today, deep neural networks (DNNs) are designed solely to improve prediction accuracy, often ignoring real-world constraints such as compute and memory resources. Embedded vision developers typically prefer to use these state-of-the-art (SOTA) DNNs from the research literature due to the costs and expertise needed to develop new models. However, these SOTA DNNs are typically too resource-hungry to be run on embedded processors. Neural architecture search (NAS) is an effective approach to bridge the gap between optimal network design and efficient deployment. In this presentation, Hiram Rayo Torres Rodriguez, Senior AI Research Engineer at NXP Semiconductors, explains the principles of NAS. He shows how NAS can enable efficient computer vision at the edge by considering deployment aspects (e.g., the efficiency of quantized operators) to derive tailor-made solutions for a given edge node, and how to address NAS scalability via smart search space design and efficient performance estimation. |
ANALYZING AND IMPLEMENTING MULTIMODAL PERCEPTION |
 |
|
Augmenting Visual AI through Radar and Camera Fusion In this presentation, Sébastien Taylor, Vice President of Research and Development for Au-Zone Technologies, discusses limitations of camera-based AI and how radar can be leveraged to address these limitations. He covers common radar data representations and how AI can be used with radar. He then explains how radar and camera data can be fused into a single sensor-fusion AI model. Taylor also dives into low-level radar data cubes and how these can provide richer data for the model while introducing some new challenges, and he discusses methods to tackle these challenges. Finally he explores some example scenarios comparing a radar-camera combination with a stereo camera to highlight the trade-offs of the two approaches. |
 |
|
Entering the Era of Multimodal Perception Humans rely on multiple senses to quickly and accurately obtain the most important information we need. Similarly, developers have begun using multiple types of sensors to improve machine perception. To date, this has mostly been done with “late fusion” approaches, in which separate ML models are trained for each type of sensor data, and the outputs of these models are combined in an ad hoc manner. However, such systems have proven difficult to implement and disappointing in their perception performance. We are now witnessing a transition away from this siloed sensor approach. Recent research shows that superior perception performance can be realized by training a single ML model on multiple types of sensor data. In this talk, Simon Morris, serial tech entrepreneur and Start-Up Advisor at Connected Vision Advisors, explains why this new approach to multimodal perception will soon dominate and outlines the key business challenges and opportunities that are emerging as a result, including challenges and opportunities in frameworks, tools, databases and models. |
UPCOMING INDUSTRY EVENTS |
|
Leveraging Synthetic Data for Real-time Visual Human Behavior Analysis Using the SKY ENGINE AI Platform – SKY ENGINE AI Webinar: September 26, 2024, 9:00 am PT 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 |
|
SiMa.ai Expands Its ONE Platform for Edge AI with MLSoC Modalix, a New Product Family for Generative AI Synaptics Partnership Enables VS680 SoC Set-top Box Support for AI-enabled Frame-accurate SDR-to-HDR Video Conversion e-con Systems, FRAMOS and Vision Components are among the Alliance Member companies participating in the upcoming VISION 2024 Show |
EDGE AI AND VISION PRODUCT OF THE YEAR WINNER SHOWCASE |
 |
|
Tenyks Data-Centric CoPilot for Vision (Best Edge AI Developer Tool) Tenyks’ Data-Centric CoPilot for Vision is the 2024 Edge AI and Vision Product of the Year Award Winner in the Edge AI Developer Tools category. The Data-Centric CoPilot for Vision platform helps computer vision teams develop production-ready models 8x faster. The platform enables machine learning (ML) teams to mine edge cases, failure patterns and annotation quality issues for more accurate, capable and robust models. In addition, it helps ML teams intelligently sub-sample datasets to increase model quality and cost efficiency. The platform supports the use of multimodal prompts to quickly compare model performance on customized training scenarios, such as pedestrians jaywalking at dusk, in order to discover blind spots and enhance reliability. ML teams can also leverage powerful search functionality to conduct data curation in hours vs. weeks. One notable feature of the platform is its multimodal Embeddings-as-a-Service (EaaS) to expertly organize, curate, and manage datasets. Another key platform feature is the streamlined cloud integration, supporting a multitude of cloud storage services and facilitating effortless access and management of large-scale datasets. Please see here for more information on Tenyks’ Data-Centric CoPilot for Vision. 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. |
