Demos, Demos & More Demos!

Real-Time Object Tracking with OpenMV

This session will demonstrate how to use MicroPython, along with a quantized TensorFlow Lite Microcontroller model, to run a cup detection program that keeps track of the number of cups in the image. You'll learn how to create a dataset, train a model and deploy the neural network model onto the OpenMV H7 board. The demo tracks the number of cups in an image by first using classical image processing methods on each frame prior to passing the interim results to the model to find blobs and make bounding boxes based on the blobs' colors, shapes, widths, and heights. (Representatives: Carlo Grisafi, IoT Developer Advocate)

Moving the Gym to Your Living Room: Body Pose Tracking on Your Smart TV

The lifestyle changes we have all experienced over the past year have brought a big focus on home fitness. At the same time, we have seen many advances in deep learning technology, along with increased camera presence in smart TVs and other home devices. In this session, we will demonstrate our fitness application performing real-time body pose estimation and tracking on Arm CPUs and GPUs. We will share some excellent results using Google’s BlazePose model, and discuss the challenges and opportunities for creating immersive experiences on Arm platforms. (Representatives: Mina Dimova, Staff Software Engineer)

An Open Source Approach To Cloud Native Vision Workload Deployment On Arm

This demonstration will showcase how to use open-source lightweight Kubernetes-based orchestration for deploying and managing vision and machine learning workloads on edge devices. (Representatives: Umair Hashim, Principal Solutions Engineer)

Color-Based Object Detection System for Visual AI Applications

BASF will demonstrate an object detection system consisting of cameras, LED lighting and colors designed to serve as product identifiers. The system can be used to improve the object detection performance for use cases involving manufactured consumer items as well as to label neural network training image sets. (Representatives: Ian Childers, Head of Technology)

Sensor Fusion + Semantic Segmentation Processing on Blaize Pathfinder P1600

This session will demonstrate sensor fusion and semantic segmentation in a use case fusing data received from full HD cameras and from lidar and radar sensors on a standalone embedded system, the Blaize Pathfinder P1600. (Representatives: Doug Watt, Director Field Applications Engineering)

Multi-Camera Object Detection on the Blaize Xplorer X1600E

This demonstration will show multi-camera object detection running on the Blaize Xplorer X1600E. The use case processes 5 independent HD video streams using 5 independent YoloV3 networks with less than 100 ms latency, for true real-time processing at the edge. (Representatives: Shawn Holiday, Sr. Director, Customer Success)

Blaize Picasso SDK: People & Pose Detection, Key Point Tracking

This session demonstrates a high resolution, multi-neural network and multi-function graph-native application built on the Blaize Picasso SDK, with the Blaize Xplorer X1600P as the host system. (Representatives: Rajesh Anantharaman, Sr. Director of Products)

Cadence Demonstration of Vision and AI Applications on Tensilica DSP-based Platforms

In this session, Amol Borkar, Senior Product Manager and Marketing of Vision and AI DSPs at Cadence, will demonstrate real-time examples of driver monitoring, automotive perception, 6DoF SLAM, and semantic segmentation, all running on Tensilica processors and DSPs. (Representatives: Amol Borkar, Senior Product Manager and Marketing of Vision and AI DSPs)

Leveraging the High Throughput of the Hailo-8 AI Processor to Improve Large-Scale Object Detection on the Edge

The high throughput delivered by the Hailo-8 AI processor can be leveraged not only to process multiple cameras at the same time but also to improve detection accuracy in high-resolution video. Tiling is a dedicated application that allows developers to break down high-resolution images into multiple input streams, thus making small and faraway objects detectable even in very busy and dynamic environments. Hailo's out-of-the-box TAPPAS applications are a great way to hit the ground running, making the development and deployment of models easier and accelerating time-to-market. (Representative: Yaniv Sulkes, VP Product Marketing & Automotive)

State-of-the-Art Object Detection on the Edge with Hailo-8 AI Processor

YoloV5m is a high-end neural network model for object detection, a fundamental computer vision task. The Hailo-8 AI chip can run this model at real-time frame rates with a high-resolution video input, providing a powerful solution for surveillance applications in Smart City, Smart Retail, Enterprise and other markets. The Hailo-8 is supported by a dedicated out-of-the-box application included in the Hailo TAPPAS toolkit, designed to accelerate time-to-market and make it easy to develop and deploy high-performance edge applications. (Representative: Nadav Eden, Technical Account Manager)

Lattice Semiconductor's CrossLink-NX: Human Presence and Counting

This demonstration showcases a development board equipped with a CrossLink-NX FPGA, based on the Lattice Semiconductor Nexus platform. The FPGA is programmed to perform human presence detection and counting, common AI use cases found in many applications. (Representative: Hussein Osman, Product Marketing Manager)

Helion ISP on ECP5

Lattice Semiconductor has partnered with Helion-Vision to bring its industry-leading IONOS image signal processing (ISP) IP portfolio to the Lattice ECP5 FPGA. This demonstration showcases the quality of Helion-Vision's image processing algorithms, along with showing optional features such as image overlay. (Representative: Mark Hoopes, Director, Product Marketing)

Ignitarium AI Unit inspection

This demonstration was developed by Lattice Semiconductor partner Ignitarium, experts in applied AI technology. It shows how you can implement efficient machine learning/AI algorithms on the Lattice ECP5 FPGA to perform complex tasks. In this case, the algorithm looks for defective products as they pass along a conveyor belt, with higher speed and better quality of results than any traditional or human-powered approach. (Representative: Hussein Osamn, Product Marketing Manager)

Spatial AI and CV for Human Machine Safety

In situations where people need to interact with potentially dangerous equipment, it is often not possible to physically guard or “dumb-guard” (e.g., with sonar or laser, for example) the equipment. In such cases, you need to discern not just the presence of an object in the equipment’s “red zone,” but also what that object is; some things (wooden boards with a saw, for example) should be there, while some things (human limbs, for example) shouldn't be! Embedded spatial artificial intelligence and computer vision, as demonstrated in this session, enables such machines to intelligently perceive objects so that they can smart-guard to protect people while not adversely impacting normal operation. (Representative: Shawn McLaughlin, Vice President)

Neural-Inference-Controlled Crop/Zoom and H.265 Encode

In this demonstration, we will show how to use a neural network to guide what portion of a high-resolution image sensor (12MP) is output to a 2MP (1920x1080) h.265-encoded video stream. This capability allows a neural network or computer vision algorithm (e.g., motion estimation) to guide where the action is in a given scene, and then zoom (6x lossless) into that action, h.265-encoding the resultant 1920x1080 region of interest. (Representative: Martin Peterlin, Chief Technology Officer)

From-Behind Collision Detection for People Who Ride Bikes

In this demonstration, we'll show an example of localizing and estimating the trajectories of vehicles behind you to determine if they are on a trajectory to hit you. Too many people have been struck from behind by distracted drivers. Let's use embedded, performant spatial artificial intelligence and computer vision to solve this problem! (Representative: Brandon Gilles, CEO)

Efficient Driver Monitoring for Automotive Cameras

In this demonstration, Nextchip, in collaboration with PathPartner Technology, will present a driver monitoring system (DMS) solution based on its ADAS SoC, APACHE4. The solution includes both distraction and drowsiness detection algorithms, ported to the CEVA XM-4 DSP (delivering 77 GMAC of performance) in the APACHE4. Thanks to the power of a well-designed platform and well-optimized algorithms, the solution delivers both high frame rates and high accuracy. (Representative: Jessie Lee, Technical Marketer(GM))

Nextchip’s Imaging Signal Processor: Supported Functions

In this session, Nextchip will showcase its imaging signal processor (ISP) expertise by demonstrating various ISP features for automotive applications. Nextchip’s ISP pipeline is developed in full by the company, leading to improved tuning capabilities. Nextchip will demonstrate the following algorithms and functions running on the ISP: high dynamic range (HDR), LED flicker mitigation (LFM), auto-calibration, and dewarp. (Representative: Jesse Lee Kim, Technical Marketer(GM))

Enabling Smart Automotive Camera ADAS SoCs

In this demonstration, Nextchip will present its image processors for the edge, APACHE4 and 5, which are ADAS SoCs empowering cameras with various “smart” functions. Because the products are designed to be located in a vehicle’s camera module, they have very low power consumption and are very small. The demonstration will show the SoCs’ various use cases, along with the well-organized APACHE5 SDK. (Representative: James Kim, Technical Marketer(Director))

Pick a Card, Any Card: Using ML-Enhanced Vision for ID and Label Tracking

Security cameras can play a valuable role in tracking the whereabouts of goods and personnel within facilities, but to do so they need to be able to reliably read information on ID cards and labels. Using a deck of standard playing cards, Perceive will demonstrate how the combination of a high-definition image sensor and the company’s Ergo edge inference processor can read information at distances beyond the capability of the human eye and expand the role of computer vision in applications such as inventory control and access management. (Representative: Kathy Cook, VP, Business Development)

Smart Video Conferencing on the Edge

Enhancements in AI processing capabilities on edge devices are enabling a richer video conferencing experience. These capabilities are leveraged in use cases that range from biometrics and automatic access of calendar and video conferencing applications through face and voice identification to a smarter framing of the scene in front of the camera, and understanding the language for auto-subtitling...all on the edge. The Synaptics VS680 and VS780 SOCs were designed with smart video conferencing use cases in mind. They are highly integrated to implement cost-optimized video conferencing devices in different form factors and boast the needed AI computation. In this demonstration, the VS680 will run smart framing, where the captured scene in front of the camera is dynamically adjusted to show the region of interest using machine learning. (Representatives: Zafer Diab, Director of Product Marketing & Xin Li, FAE Engineer)

Real-Time Video Post-processing Using Machine Learning

Enhancements in AI processing capabilities on edge devices are enabling significant enhancements in video scaling and post-processing versus, compared to what had been possible with traditional scaling integrated in SoCs. These enhancements enable the scaling and post-processing to be theme-based. Scaling for sports content can be adjusted for the high motion content, for example, while scaling for video conferencing can be optimized for the (mostly) static and moving face content. Synaptics SoCs integrate an internally developed machine learning engine called QDEO.ai, which performs super-resolution scaling that takes the theme of the video into consideration. This demonstration will show a side-by-side comparison of scaling using the traditional hardware scaler and the QDEO.ai scaled video. Quality enhancements are accentuated when performed on lower-bitrate input video, such as with videoconferencing. (Representatives: Zafer Diab, Director of Product Marketing & Xin Li, FAE Engineer)

From Neural Network to Edge with Synaptics

Synaptics will present an overview of the Katana Edge AI processor and tensaicc, the compiler that Eta Compute is developing for use with Katana. After we introduce Katana's novel multicore architecture, you'll see a live demonstration of how tensaicc compiles a neural network and generates power-, cycle- and memory-optimized code that takes advantage of the architecture. (Representative: Vineet Ganju, VP/GM Audio Business Unit)

Simultaneous Localization and Mapping (SLAM) Acceleration on DesignWare ARC EV7x Processors

Simultaneous Localization and Mapping (SLAM) creates and updates a map of an unknown environment while at the same time keeping track of an agent's location within it. SLAM is a key component in a variety of systems, such as autonomous vehicles, robotics, and augmented and virtual reality. This demonstration shows how to accelerate a SLAM software stack by offloading some of the processing to the ARC EV7x processor, as well as how adding an extra core to parallelize the algorithms can further increase the acceleration. We’ll also show how, with an EV7x-based system, you can incorporate a deep neural network engine to expand the system's intelligence. (Representative: Liliya Tazieva, Software Engineer)

SRGAN Super Resolution on DesignWare ARC EV7x Processors

Image super-resolution techniques reconstruct a higher-resolution image from a lower-resolution one. Although this can be done with classical vision algorithms, the results are inferior to what neural-network-based solutions can now offer. In this demo, we show how a Generative Adversarial Network can be used to intelligently infer the missing pixel data and generate a high-quality high-resolution image. The demo will run on the ARC EV7x processor and showcase the Deep Neural Network accelerator engine. (Representative: Liliya Tazieva, Software Engineer)

X-Ray Classification Solution for COVID-19 and Pneumonia Detection

Spline.AI is collaborating with Amazon Web Services and Xilinx to deliver an open-source, open-model, X-Ray classification solution for COVID-19 and pneumonia detection. This demonstration will show the model, deployed on the Xilinx Zynq UltraScale+- MPSoC device-based ZCU104 evaluation kit and leveraging the Xilinx deep learning processor unit (DPU), a soft-IP tensor accelerator that is powerful enough to run a variety of neural networks, including classification and detection of diseases. (Representative: Quenton Hall, AI Architect)

Out-of-the-Box with the Kria KV260 Starter Kit: Up and Running in Under an Hour

This demonstration will provide a detailed look at the Kria KV260 Vision AI Starter Kit and its companion basic accessory pack. We’ll show you just how quickly and easily you can get our Smart Camera accelerated application up and running, with no FPGA experience required. (Representative: Karan Kantharia, Product Line Manager)

Visual Machine Learning and Natural Language Processing Fusion on the Xilinx Kria SOM

This demonstration shows the new  Kria K26 SOM running both vision machine learning (ML) and natural language processing on the same Xilinx Zynq-based platform. This solution showcases the dynamic switching capabilities of the Xilinx deep learning processor unit and integrates audio ML keyword-spotting to control the video display. (Representative: Girish Malipeddi, Director of Video and Imaging Solutions)

Real-Time Object Tracking with OpenMV

This session will demonstrate how to use MicroPython, along with a quantized TensorFlow Lite Microcontroller model, to run a cup detection program that keeps track of the number of cups in the image. You'll learn how to create a dataset, train a model and deploy the neural network model onto the OpenMV H7 board. The demo tracks the number of cups in an image by first using classical image processing methods on each frame prior to passing the interim results to the model to find blobs and make bounding boxes based on the blobs' colors, shapes, widths, and heights. (Representatives: Carlo Grisafi, IoT Developer Advocate)

Moving the Gym to Your Living Room: Body Pose Tracking on Your Smart TV

The lifestyle changes we have all experienced over the past year have brought a big focus on home fitness. At the same time, we have seen many advances in deep learning technology, along with increased camera presence in smart TVs and other home devices. In this session, we will demonstrate our fitness application performing real-time body pose estimation and tracking on Arm CPUs and GPUs. We will share some excellent results using Google’s BlazePose model, and discuss the challenges and opportunities for creating immersive experiences on Arm platforms. (Representatives: Mina Dimova, Staff Software Engineer)

An Open Source Approach To Cloud Native Vision Workload Deployment On Arm

This demonstration will showcase how to use open-source lightweight Kubernetes-based orchestration for deploying and managing vision and machine learning workloads on edge devices. (Representatives: Umair Hashim, Principal Solutions Engineer)

Color-Based Object Detection System for Visual AI Applications

BASF will demonstrate an object detection system consisting of cameras, LED lighting and colors designed to serve as product identifiers. The system can be used to improve the object detection performance for use cases involving manufactured consumer items as well as to label neural network training image sets. (Representatives: Ian Childers, Head of Technology)

Sensor Fusion + Semantic Segmentation Processing on Blaize Pathfinder P1600

This session will demonstrate sensor fusion and semantic segmentation in a use case fusing data received from full HD cameras and from lidar and radar sensors on a standalone embedded system, the Blaize Pathfinder P1600. (Representatives: Doug Watt, Director Field Applications Engineering)

Multi-Camera Object Detection on the Blaize Xplorer X1600E

This demonstration will show multi-camera object detection running on the Blaize Xplorer X1600E. The use case processes 5 independent HD video streams using 5 independent YoloV3 networks with less than 100 ms latency, for true real-time processing at the edge. (Representatives: Shawn Holiday, Sr. Director, Customer Success)

Blaize Picasso SDK: People & Pose Detection, Key Point Tracking

This session demonstrates a high resolution, multi-neural network and multi-function graph-native application built on the Blaize Picasso SDK, with the Blaize Xplorer X1600P as the host system. (Representatives: Rajesh Anantharaman, Sr. Director of Products)

Cadence Demonstration of Vision and AI Applications on Tensilica DSP-based Platforms

In this session, Amol Borkar, Senior Product Manager and Marketing of Vision and AI DSPs at Cadence, will demonstrate real-time examples of driver monitoring, automotive perception, 6DoF SLAM, and semantic segmentation, all running on Tensilica processors and DSPs. (Representatives: Amol Borkar, Senior Product Manager and Marketing of Vision and AI DSPs)

Leveraging the High Throughput of the Hailo-8 AI Processor to Improve Large-Scale Object Detection on the Edge

The high throughput delivered by the Hailo-8 AI processor can be leveraged not only to process multiple cameras at the same time but also to improve detection accuracy in high-resolution video. Tiling is a dedicated application that allows developers to break down high-resolution images into multiple input streams, thus making small and faraway objects detectable even in very busy and dynamic environments. Hailo's out-of-the-box TAPPAS applications are a great way to hit the ground running, making the development and deployment of models easier and accelerating time-to-market. (Representative: Yaniv Sulkes, VP Product Marketing & Automotive)

State-of-the-Art Object Detection on the Edge with Hailo-8 AI Processor

YoloV5m is a high-end neural network model for object detection, a fundamental computer vision task. The Hailo-8 AI chip can run this model at real-time frame rates with a high-resolution video input, providing a powerful solution for surveillance applications in Smart City, Smart Retail, Enterprise and other markets. The Hailo-8 is supported by a dedicated out-of-the-box application included in the Hailo TAPPAS toolkit, designed to accelerate time-to-market and make it easy to develop and deploy high-performance edge applications. (Representative: Nadav Eden, Technical Account Manager)

Lattice Semiconductor's CrossLink-NX: Human Presence and Counting

This demonstration showcases a development board equipped with a CrossLink-NX FPGA, based on the Lattice Semiconductor Nexus platform. The FPGA is programmed to perform human presence detection and counting, common AI use cases found in many applications. (Representative: Hussein Osman, Product Marketing Manager)

Helion ISP on ECP5

Lattice Semiconductor has partnered with Helion-Vision to bring its industry-leading IONOS image signal processing (ISP) IP portfolio to the Lattice ECP5 FPGA. This demonstration showcases the quality of Helion-Vision's image processing algorithms, along with showing optional features such as image overlay. (Representative: Mark Hoopes, Director, Product Marketing)

Ignitarium AI Unit inspection

This demonstration was developed by Lattice Semiconductor partner Ignitarium, experts in applied AI technology. It shows how you can implement efficient machine learning/AI algorithms on the Lattice ECP5 FPGA to perform complex tasks. In this case, the algorithm looks for defective products as they pass along a conveyor belt, with higher speed and better quality of results than any traditional or human-powered approach. (Representative: Hussein Osamn, Product Marketing Manager)

Spatial AI and CV for Human Machine Safety

In situations where people need to interact with potentially dangerous equipment, it is often not possible to physically guard or “dumb-guard” (e.g., with sonar or laser, for example) the equipment. In such cases, you need to discern not just the presence of an object in the equipment’s “red zone,” but also what that object is; some things (wooden boards with a saw, for example) should be there, while some things (human limbs, for example) shouldn't be! Embedded spatial artificial intelligence and computer vision, as demonstrated in this session, enables such machines to intelligently perceive objects so that they can smart-guard to protect people while not adversely impacting normal operation. (Representative: Shawn McLaughlin, Vice President)

Neural-Inference-Controlled Crop/Zoom and H.265 Encode

In this demonstration, we will show how to use a neural network to guide what portion of a high-resolution image sensor (12MP) is output to a 2MP (1920x1080) h.265-encoded video stream. This capability allows a neural network or computer vision algorithm (e.g., motion estimation) to guide where the action is in a given scene, and then zoom (6x lossless) into that action, h.265-encoding the resultant 1920x1080 region of interest. (Representative: Martin Peterlin, Chief Technology Officer)

From-Behind Collision Detection for People Who Ride Bikes

In this demonstration, we'll show an example of localizing and estimating the trajectories of vehicles behind you to determine if they are on a trajectory to hit you. Too many people have been struck from behind by distracted drivers. Let's use embedded, performant spatial artificial intelligence and computer vision to solve this problem! (Representative: Brandon Gilles, CEO)

Efficient Driver Monitoring for Automotive Cameras

In this demonstration, Nextchip, in collaboration with PathPartner Technology, will present a driver monitoring system (DMS) solution based on its ADAS SoC, APACHE4. The solution includes both distraction and drowsiness detection algorithms, ported to the CEVA XM-4 DSP (delivering 77 GMAC of performance) in the APACHE4. Thanks to the power of a well-designed platform and well-optimized algorithms, the solution delivers both high frame rates and high accuracy. (Representative: Jessie Lee, Technical Marketer(GM))

Nextchip’s Imaging Signal Processor: Supported Functions

In this session, Nextchip will showcase its imaging signal processor (ISP) expertise by demonstrating various ISP features for automotive applications. Nextchip’s ISP pipeline is developed in full by the company, leading to improved tuning capabilities. Nextchip will demonstrate the following algorithms and functions running on the ISP: high dynamic range (HDR), LED flicker mitigation (LFM), auto-calibration, and dewarp. (Representative: Jesse Lee Kim, Technical Marketer(GM))

Enabling Smart Automotive Camera ADAS SoCs

In this demonstration, Nextchip will present its image processors for the edge, APACHE4 and 5, which are ADAS SoCs empowering cameras with various “smart” functions. Because the products are designed to be located in a vehicle’s camera module, they have very low power consumption and are very small. The demonstration will show the SoCs’ various use cases, along with the well-organized APACHE5 SDK. (Representative: James Kim, Technical Marketer(Director))

Pick a Card, Any Card: Using ML-Enhanced Vision for ID and Label Tracking

Security cameras can play a valuable role in tracking the whereabouts of goods and personnel within facilities, but to do so they need to be able to reliably read information on ID cards and labels. Using a deck of standard playing cards, Perceive will demonstrate how the combination of a high-definition image sensor and the company’s Ergo edge inference processor can read information at distances beyond the capability of the human eye and expand the role of computer vision in applications such as inventory control and access management. (Representative: Kathy Cook, VP, Business Development)

Smart Video Conferencing on the Edge

Enhancements in AI processing capabilities on edge devices are enabling a richer video conferencing experience. These capabilities are leveraged in use cases that range from biometrics and automatic access of calendar and video conferencing applications through face and voice identification to a smarter framing of the scene in front of the camera, and understanding the language for auto-subtitling...all on the edge. The Synaptics VS680 and VS780 SOCs were designed with smart video conferencing use cases in mind. They are highly integrated to implement cost-optimized video conferencing devices in different form factors and boast the needed AI computation. In this demonstration, the VS680 will run smart framing, where the captured scene in front of the camera is dynamically adjusted to show the region of interest using machine learning. (Representatives: Zafer Diab, Director of Product Marketing & Xin Li, FAE Engineer)

Real-Time Video Post-processing Using Machine Learning

Enhancements in AI processing capabilities on edge devices are enabling significant enhancements in video scaling and post-processing versus, compared to what had been possible with traditional scaling integrated in SoCs. These enhancements enable the scaling and post-processing to be theme-based. Scaling for sports content can be adjusted for the high motion content, for example, while scaling for video conferencing can be optimized for the (mostly) static and moving face content. Synaptics SoCs integrate an internally developed machine learning engine called QDEO.ai, which performs super-resolution scaling that takes the theme of the video into consideration. This demonstration will show a side-by-side comparison of scaling using the traditional hardware scaler and the QDEO.ai scaled video. Quality enhancements are accentuated when performed on lower-bitrate input video, such as with videoconferencing. (Representatives: Zafer Diab, Director of Product Marketing & Xin Li, FAE Engineer)

From Neural Network to Edge with Synaptics

Synaptics will present an overview of the Katana Edge AI processor and tensaicc, the compiler that Eta Compute is developing for use with Katana. After we introduce Katana's novel multicore architecture, you'll see a live demonstration of how tensaicc compiles a neural network and generates power-, cycle- and memory-optimized code that takes advantage of the architecture. (Representative: Vineet Ganju, VP/GM Audio Business Unit)

Simultaneous Localization and Mapping (SLAM) Acceleration on DesignWare ARC EV7x Processors

Simultaneous Localization and Mapping (SLAM) creates and updates a map of an unknown environment while at the same time keeping track of an agent's location within it. SLAM is a key component in a variety of systems, such as autonomous vehicles, robotics, and augmented and virtual reality. This demonstration shows how to accelerate a SLAM software stack by offloading some of the processing to the ARC EV7x processor, as well as how adding an extra core to parallelize the algorithms can further increase the acceleration. We’ll also show how, with an EV7x-based system, you can incorporate a deep neural network engine to expand the system's intelligence. (Representative: Liliya Tazieva, Software Engineer)

SRGAN Super Resolution on DesignWare ARC EV7x Processors

Image super-resolution techniques reconstruct a higher-resolution image from a lower-resolution one. Although this can be done with classical vision algorithms, the results are inferior to what neural-network-based solutions can now offer. In this demo, we show how a Generative Adversarial Network can be used to intelligently infer the missing pixel data and generate a high-quality high-resolution image. The demo will run on the ARC EV7x processor and showcase the Deep Neural Network accelerator engine. (Representative: Liliya Tazieva, Software Engineer)

X-Ray Classification Solution for COVID-19 and Pneumonia Detection

Spline.AI is collaborating with Amazon Web Services and Xilinx to deliver an open-source, open-model, X-Ray classification solution for COVID-19 and pneumonia detection. This demonstration will show the model, deployed on the Xilinx Zynq UltraScale+- MPSoC device-based ZCU104 evaluation kit and leveraging the Xilinx deep learning processor unit (DPU), a soft-IP tensor accelerator that is powerful enough to run a variety of neural networks, including classification and detection of diseases. (Representative: Quenton Hall, AI Architect)

Out-of-the-Box with the Kria KV260 Starter Kit: Up and Running in Under an Hour

This demonstration will provide a detailed look at the Kria KV260 Vision AI Starter Kit and its companion basic accessory pack. We’ll show you just how quickly and easily you can get our Smart Camera accelerated application up and running, with no FPGA experience required. (Representative: Karan Kantharia, Product Line Manager)

Visual Machine Learning and Natural Language Processing Fusion on the Xilinx Kria SOM

This demonstration shows the new  Kria K26 SOM running both vision machine learning (ML) and natural language processing on the same Xilinx Zynq-based platform. This solution showcases the dynamic switching capabilities of the Xilinx deep learning processor unit and integrates audio ML keyword-spotting to control the video display. (Representative: Girish Malipeddi, Director of Video and Imaging Solutions)

Real-Time Object Tracking with OpenMV

This session will demonstrate how to use MicroPython, along with a quantized TensorFlow Lite Microcontroller model, to run a cup detection program that keeps track of the number of cups in the image. You'll learn how to create a dataset, train a model and deploy the neural network model onto the OpenMV H7 board. The demo tracks the number of cups in an image by first using classical image processing methods on each frame prior to passing the interim results to the model to find blobs and make bounding boxes based on the blobs' colors, shapes, widths, and heights. (Representatives: Carlo Grisafi, IoT Developer Advocate)

Moving the Gym to Your Living Room: Body Pose Tracking on Your Smart TV

The lifestyle changes we have all experienced over the past year have brought a big focus on home fitness. At the same time, we have seen many advances in deep learning technology, along with increased camera presence in smart TVs and other home devices. In this session, we will demonstrate our fitness application performing real-time body pose estimation and tracking on Arm CPUs and GPUs. We will share some excellent results using Google’s BlazePose model, and discuss the challenges and opportunities for creating immersive experiences on Arm platforms. (Representatives: Mina Dimova, Staff Software Engineer)

An Open Source Approach To Cloud Native Vision Workload Deployment On Arm

This demonstration will showcase how to use open-source lightweight Kubernetes-based orchestration for deploying and managing vision and machine learning workloads on edge devices. (Representatives: Umair Hashim, Principal Solutions Engineer)

Color-Based Object Detection System for Visual AI Applications

BASF will demonstrate an object detection system consisting of cameras, LED lighting and colors designed to serve as product identifiers. The system can be used to improve the object detection performance for use cases involving manufactured consumer items as well as to label neural network training image sets. (Representatives: Ian Childers, Head of Technology)

Sensor Fusion + Semantic Segmentation Processing on Blaize Pathfinder P1600

This session will demonstrate sensor fusion and semantic segmentation in a use case fusing data received from full HD cameras and from lidar and radar sensors on a standalone embedded system, the Blaize Pathfinder P1600. (Representatives: Doug Watt, Director Field Applications Engineering)

Multi-Camera Object Detection on the Blaize Xplorer X1600E

This demonstration will show multi-camera object detection running on the Blaize Xplorer X1600E. The use case processes 5 independent HD video streams using 5 independent YoloV3 networks with less than 100 ms latency, for true real-time processing at the edge. (Representatives: Shawn Holiday, Sr. Director, Customer Success)

Blaize Picasso SDK: People & Pose Detection, Key Point Tracking

This session demonstrates a high resolution, multi-neural network and multi-function graph-native application built on the Blaize Picasso SDK, with the Blaize Xplorer X1600P as the host system. (Representatives: Rajesh Anantharaman, Sr. Director of Products)

Cadence Demonstration of Vision and AI Applications on Tensilica DSP-based Platforms

In this session, Amol Borkar, Senior Product Manager and Marketing of Vision and AI DSPs at Cadence, will demonstrate real-time examples of driver monitoring, automotive perception, 6DoF SLAM, and semantic segmentation, all running on Tensilica processors and DSPs. (Representatives: Amol Borkar, Senior Product Manager and Marketing of Vision and AI DSPs)

Leveraging the High Throughput of the Hailo-8 AI Processor to Improve Large-Scale Object Detection on the Edge

The high throughput delivered by the Hailo-8 AI processor can be leveraged not only to process multiple cameras at the same time but also to improve detection accuracy in high-resolution video. Tiling is a dedicated application that allows developers to break down high-resolution images into multiple input streams, thus making small and faraway objects detectable even in very busy and dynamic environments. Hailo's out-of-the-box TAPPAS applications are a great way to hit the ground running, making the development and deployment of models easier and accelerating time-to-market. (Representative: Yaniv Sulkes, VP Product Marketing & Automotive)

State-of-the-Art Object Detection on the Edge with Hailo-8 AI Processor

YoloV5m is a high-end neural network model for object detection, a fundamental computer vision task. The Hailo-8 AI chip can run this model at real-time frame rates with a high-resolution video input, providing a powerful solution for surveillance applications in Smart City, Smart Retail, Enterprise and other markets. The Hailo-8 is supported by a dedicated out-of-the-box application included in the Hailo TAPPAS toolkit, designed to accelerate time-to-market and make it easy to develop and deploy high-performance edge applications. (Representative: Nadav Eden, Technical Account Manager)

Lattice Semiconductor's CrossLink-NX: Human Presence and Counting

This demonstration showcases a development board equipped with a CrossLink-NX FPGA, based on the Lattice Semiconductor Nexus platform. The FPGA is programmed to perform human presence detection and counting, common AI use cases found in many applications. (Representative: Hussein Osman, Product Marketing Manager)

Helion ISP on ECP5

Lattice Semiconductor has partnered with Helion-Vision to bring its industry-leading IONOS image signal processing (ISP) IP portfolio to the Lattice ECP5 FPGA. This demonstration showcases the quality of Helion-Vision's image processing algorithms, along with showing optional features such as image overlay. (Representative: Mark Hoopes, Director, Product Marketing)

Ignitarium AI Unit inspection

This demonstration was developed by Lattice Semiconductor partner Ignitarium, experts in applied AI technology. It shows how you can implement efficient machine learning/AI algorithms on the Lattice ECP5 FPGA to perform complex tasks. In this case, the algorithm looks for defective products as they pass along a conveyor belt, with higher speed and better quality of results than any traditional or human-powered approach. (Representative: Hussein Osamn, Product Marketing Manager)

Spatial AI and CV for Human Machine Safety

In situations where people need to interact with potentially dangerous equipment, it is often not possible to physically guard or “dumb-guard” (e.g., with sonar or laser, for example) the equipment. In such cases, you need to discern not just the presence of an object in the equipment’s “red zone,” but also what that object is; some things (wooden boards with a saw, for example) should be there, while some things (human limbs, for example) shouldn't be! Embedded spatial artificial intelligence and computer vision, as demonstrated in this session, enables such machines to intelligently perceive objects so that they can smart-guard to protect people while not adversely impacting normal operation. (Representative: Shawn McLaughlin, Vice President)

Neural-Inference-Controlled Crop/Zoom and H.265 Encode

In this demonstration, we will show how to use a neural network to guide what portion of a high-resolution image sensor (12MP) is output to a 2MP (1920x1080) h.265-encoded video stream. This capability allows a neural network or computer vision algorithm (e.g., motion estimation) to guide where the action is in a given scene, and then zoom (6x lossless) into that action, h.265-encoding the resultant 1920x1080 region of interest. (Representative: Martin Peterlin, Chief Technology Officer)

From-Behind Collision Detection for People Who Ride Bikes

In this demonstration, we'll show an example of localizing and estimating the trajectories of vehicles behind you to determine if they are on a trajectory to hit you. Too many people have been struck from behind by distracted drivers. Let's use embedded, performant spatial artificial intelligence and computer vision to solve this problem! (Representative: Brandon Gilles, CEO)

Efficient Driver Monitoring for Automotive Cameras

In this demonstration, Nextchip, in collaboration with PathPartner Technology, will present a driver monitoring system (DMS) solution based on its ADAS SoC, APACHE4. The solution includes both distraction and drowsiness detection algorithms, ported to the CEVA XM-4 DSP (delivering 77 GMAC of performance) in the APACHE4. Thanks to the power of a well-designed platform and well-optimized algorithms, the solution delivers both high frame rates and high accuracy. (Representative: Jessie Lee, Technical Marketer(GM))

Nextchip’s Imaging Signal Processor: Supported Functions

In this session, Nextchip will showcase its imaging signal processor (ISP) expertise by demonstrating various ISP features for automotive applications. Nextchip’s ISP pipeline is developed in full by the company, leading to improved tuning capabilities. Nextchip will demonstrate the following algorithms and functions running on the ISP: high dynamic range (HDR), LED flicker mitigation (LFM), auto-calibration, and dewarp. (Representative: Jesse Lee Kim, Technical Marketer(GM))

Enabling Smart Automotive Camera ADAS SoCs

In this demonstration, Nextchip will present its image processors for the edge, APACHE4 and 5, which are ADAS SoCs empowering cameras with various “smart” functions. Because the products are designed to be located in a vehicle’s camera module, they have very low power consumption and are very small. The demonstration will show the SoCs’ various use cases, along with the well-organized APACHE5 SDK. (Representative: James Kim, Technical Marketer(Director))

Pick a Card, Any Card: Using ML-Enhanced Vision for ID and Label Tracking

Security cameras can play a valuable role in tracking the whereabouts of goods and personnel within facilities, but to do so they need to be able to reliably read information on ID cards and labels. Using a deck of standard playing cards, Perceive will demonstrate how the combination of a high-definition image sensor and the company’s Ergo edge inference processor can read information at distances beyond the capability of the human eye and expand the role of computer vision in applications such as inventory control and access management. (Representative: Kathy Cook, VP, Business Development)

Smart Video Conferencing on the Edge

Enhancements in AI processing capabilities on edge devices are enabling a richer video conferencing experience. These capabilities are leveraged in use cases that range from biometrics and automatic access of calendar and video conferencing applications through face and voice identification to a smarter framing of the scene in front of the camera, and understanding the language for auto-subtitling...all on the edge. The Synaptics VS680 and VS780 SOCs were designed with smart video conferencing use cases in mind. They are highly integrated to implement cost-optimized video conferencing devices in different form factors and boast the needed AI computation. In this demonstration, the VS680 will run smart framing, where the captured scene in front of the camera is dynamically adjusted to show the region of interest using machine learning. (Representatives: Zafer Diab, Director of Product Marketing & Xin Li, FAE Engineer)

Real-Time Video Post-processing Using Machine Learning

Enhancements in AI processing capabilities on edge devices are enabling significant enhancements in video scaling and post-processing versus, compared to what had been possible with traditional scaling integrated in SoCs. These enhancements enable the scaling and post-processing to be theme-based. Scaling for sports content can be adjusted for the high motion content, for example, while scaling for video conferencing can be optimized for the (mostly) static and moving face content. Synaptics SoCs integrate an internally developed machine learning engine called QDEO.ai, which performs super-resolution scaling that takes the theme of the video into consideration. This demonstration will show a side-by-side comparison of scaling using the traditional hardware scaler and the QDEO.ai scaled video. Quality enhancements are accentuated when performed on lower-bitrate input video, such as with videoconferencing. (Representatives: Zafer Diab, Director of Product Marketing & Xin Li, FAE Engineer)

From Neural Network to Edge with Synaptics

Synaptics will present an overview of the Katana Edge AI processor and tensaicc, the compiler that Eta Compute is developing for use with Katana. After we introduce Katana's novel multicore architecture, you'll see a live demonstration of how tensaicc compiles a neural network and generates power-, cycle- and memory-optimized code that takes advantage of the architecture. (Representative: Vineet Ganju, VP/GM Audio Business Unit)

Simultaneous Localization and Mapping (SLAM) Acceleration on DesignWare ARC EV7x Processors

Simultaneous Localization and Mapping (SLAM) creates and updates a map of an unknown environment while at the same time keeping track of an agent's location within it. SLAM is a key component in a variety of systems, such as autonomous vehicles, robotics, and augmented and virtual reality. This demonstration shows how to accelerate a SLAM software stack by offloading some of the processing to the ARC EV7x processor, as well as how adding an extra core to parallelize the algorithms can further increase the acceleration. We’ll also show how, with an EV7x-based system, you can incorporate a deep neural network engine to expand the system's intelligence. (Representative: Liliya Tazieva, Software Engineer)

SRGAN Super Resolution on DesignWare ARC EV7x Processors

Image super-resolution techniques reconstruct a higher-resolution image from a lower-resolution one. Although this can be done with classical vision algorithms, the results are inferior to what neural-network-based solutions can now offer. In this demo, we show how a Generative Adversarial Network can be used to intelligently infer the missing pixel data and generate a high-quality high-resolution image. The demo will run on the ARC EV7x processor and showcase the Deep Neural Network accelerator engine. (Representative: Liliya Tazieva, Software Engineer)

X-Ray Classification Solution for COVID-19 and Pneumonia Detection

Spline.AI is collaborating with Amazon Web Services and Xilinx to deliver an open-source, open-model, X-Ray classification solution for COVID-19 and pneumonia detection. This demonstration will show the model, deployed on the Xilinx Zynq UltraScale+- MPSoC device-based ZCU104 evaluation kit and leveraging the Xilinx deep learning processor unit (DPU), a soft-IP tensor accelerator that is powerful enough to run a variety of neural networks, including classification and detection of diseases. (Representative: Quenton Hall, AI Architect)

Out-of-the-Box with the Kria KV260 Starter Kit: Up and Running in Under an Hour

This demonstration will provide a detailed look at the Kria KV260 Vision AI Starter Kit and its companion basic accessory pack. We’ll show you just how quickly and easily you can get our Smart Camera accelerated application up and running, with no FPGA experience required. (Representative: Karan Kantharia, Product Line Manager)

Visual Machine Learning and Natural Language Processing Fusion on the Xilinx Kria SOM

This demonstration shows the new  Kria K26 SOM running both vision machine learning (ML) and natural language processing on the same Xilinx Zynq-based platform. This solution showcases the dynamic switching capabilities of the Xilinx deep learning processor unit and integrates audio ML keyword-spotting to control the video display. (Representative: Girish Malipeddi, Director of Video and Imaging Solutions)

Real-Time Object Tracking with OpenMV

This session will demonstrate how to use MicroPython, along with a quantized TensorFlow Lite Microcontroller model, to run a cup detection program that keeps track of the number of cups in the image. You'll learn how to create a dataset, train a model and deploy the neural network model onto the OpenMV H7 board. The demo tracks the number of cups in an image by first using classical image processing methods on each frame prior to passing the interim results to the model to find blobs and make bounding boxes based on the blobs' colors, shapes, widths, and heights. (Representatives: Carlo Grisafi, IoT Developer Advocate)

Moving the Gym to Your Living Room: Body Pose Tracking on Your Smart TV

The lifestyle changes we have all experienced over the past year have brought a big focus on home fitness. At the same time, we have seen many advances in deep learning technology, along with increased camera presence in smart TVs and other home devices. In this session, we will demonstrate our fitness application performing real-time body pose estimation and tracking on Arm CPUs and GPUs. We will share some excellent results using Google’s BlazePose model, and discuss the challenges and opportunities for creating immersive experiences on Arm platforms. (Representatives: Mina Dimova, Staff Software Engineer)

An Open Source Approach To Cloud Native Vision Workload Deployment On Arm

This demonstration will showcase how to use open-source lightweight Kubernetes-based orchestration for deploying and managing vision and machine learning workloads on edge devices. (Representatives: Umair Hashim, Principal Solutions Engineer)

Color-Based Object Detection System for Visual AI Applications

BASF will demonstrate an object detection system consisting of cameras, LED lighting and colors designed to serve as product identifiers. The system can be used to improve the object detection performance for use cases involving manufactured consumer items as well as to label neural network training image sets. (Representatives: Ian Childers, Head of Technology)

Sensor Fusion + Semantic Segmentation Processing on Blaize Pathfinder P1600

This session will demonstrate sensor fusion and semantic segmentation in a use case fusing data received from full HD cameras and from lidar and radar sensors on a standalone embedded system, the Blaize Pathfinder P1600. (Representatives: Doug Watt, Director Field Applications Engineering)

Multi-Camera Object Detection on the Blaize Xplorer X1600E

This demonstration will show multi-camera object detection running on the Blaize Xplorer X1600E. The use case processes 5 independent HD video streams using 5 independent YoloV3 networks with less than 100 ms latency, for true real-time processing at the edge. (Representatives: Shawn Holiday, Sr. Director, Customer Success)

Blaize Picasso SDK: People & Pose Detection, Key Point Tracking

This session demonstrates a high resolution, multi-neural network and multi-function graph-native application built on the Blaize Picasso SDK, with the Blaize Xplorer X1600P as the host system. (Representatives: Rajesh Anantharaman, Sr. Director of Products)

Cadence Demonstration of Vision and AI Applications on Tensilica DSP-based Platforms

In this session, Amol Borkar, Senior Product Manager and Marketing of Vision and AI DSPs at Cadence, will demonstrate real-time examples of driver monitoring, automotive perception, 6DoF SLAM, and semantic segmentation, all running on Tensilica processors and DSPs. (Representatives: Amol Borkar, Senior Product Manager and Marketing of Vision and AI DSPs)

Leveraging the High Throughput of the Hailo-8 AI Processor to Improve Large-Scale Object Detection on the Edge

The high throughput delivered by the Hailo-8 AI processor can be leveraged not only to process multiple cameras at the same time but also to improve detection accuracy in high-resolution video. Tiling is a dedicated application that allows developers to break down high-resolution images into multiple input streams, thus making small and faraway objects detectable even in very busy and dynamic environments. Hailo's out-of-the-box TAPPAS applications are a great way to hit the ground running, making the development and deployment of models easier and accelerating time-to-market. (Representative: Yaniv Sulkes, VP Product Marketing & Automotive)

State-of-the-Art Object Detection on the Edge with Hailo-8 AI Processor

YoloV5m is a high-end neural network model for object detection, a fundamental computer vision task. The Hailo-8 AI chip can run this model at real-time frame rates with a high-resolution video input, providing a powerful solution for surveillance applications in Smart City, Smart Retail, Enterprise and other markets. The Hailo-8 is supported by a dedicated out-of-the-box application included in the Hailo TAPPAS toolkit, designed to accelerate time-to-market and make it easy to develop and deploy high-performance edge applications. (Representative: Nadav Eden, Technical Account Manager)

Lattice Semiconductor's CrossLink-NX: Human Presence and Counting

This demonstration showcases a development board equipped with a CrossLink-NX FPGA, based on the Lattice Semiconductor Nexus platform. The FPGA is programmed to perform human presence detection and counting, common AI use cases found in many applications. (Representative: Hussein Osman, Product Marketing Manager)

Helion ISP on ECP5

Lattice Semiconductor has partnered with Helion-Vision to bring its industry-leading IONOS image signal processing (ISP) IP portfolio to the Lattice ECP5 FPGA. This demonstration showcases the quality of Helion-Vision's image processing algorithms, along with showing optional features such as image overlay. (Representative: Mark Hoopes, Director, Product Marketing)

Ignitarium AI Unit inspection

This demonstration was developed by Lattice Semiconductor partner Ignitarium, experts in applied AI technology. It shows how you can implement efficient machine learning/AI algorithms on the Lattice ECP5 FPGA to perform complex tasks. In this case, the algorithm looks for defective products as they pass along a conveyor belt, with higher speed and better quality of results than any traditional or human-powered approach. (Representative: Hussein Osamn, Product Marketing Manager)

Spatial AI and CV for Human Machine Safety

In situations where people need to interact with potentially dangerous equipment, it is often not possible to physically guard or “dumb-guard” (e.g., with sonar or laser, for example) the equipment. In such cases, you need to discern not just the presence of an object in the equipment’s “red zone,” but also what that object is; some things (wooden boards with a saw, for example) should be there, while some things (human limbs, for example) shouldn't be! Embedded spatial artificial intelligence and computer vision, as demonstrated in this session, enables such machines to intelligently perceive objects so that they can smart-guard to protect people while not adversely impacting normal operation. (Representative: Shawn McLaughlin, Vice President)

Neural-Inference-Controlled Crop/Zoom and H.265 Encode

In this demonstration, we will show how to use a neural network to guide what portion of a high-resolution image sensor (12MP) is output to a 2MP (1920x1080) h.265-encoded video stream. This capability allows a neural network or computer vision algorithm (e.g., motion estimation) to guide where the action is in a given scene, and then zoom (6x lossless) into that action, h.265-encoding the resultant 1920x1080 region of interest. (Representative: Martin Peterlin, Chief Technology Officer)

From-Behind Collision Detection for People Who Ride Bikes

In this demonstration, we'll show an example of localizing and estimating the trajectories of vehicles behind you to determine if they are on a trajectory to hit you. Too many people have been struck from behind by distracted drivers. Let's use embedded, performant spatial artificial intelligence and computer vision to solve this problem! (Representative: Brandon Gilles, CEO)

Efficient Driver Monitoring for Automotive Cameras

In this demonstration, Nextchip, in collaboration with PathPartner Technology, will present a driver monitoring system (DMS) solution based on its ADAS SoC, APACHE4. The solution includes both distraction and drowsiness detection algorithms, ported to the CEVA XM-4 DSP (delivering 77 GMAC of performance) in the APACHE4. Thanks to the power of a well-designed platform and well-optimized algorithms, the solution delivers both high frame rates and high accuracy. (Representative: Jessie Lee, Technical Marketer(GM))

Nextchip’s Imaging Signal Processor: Supported Functions

In this session, Nextchip will showcase its imaging signal processor (ISP) expertise by demonstrating various ISP features for automotive applications. Nextchip’s ISP pipeline is developed in full by the company, leading to improved tuning capabilities. Nextchip will demonstrate the following algorithms and functions running on the ISP: high dynamic range (HDR), LED flicker mitigation (LFM), auto-calibration, and dewarp. (Representative: Jesse Lee Kim, Technical Marketer(GM))

Enabling Smart Automotive Camera ADAS SoCs

In this demonstration, Nextchip will present its image processors for the edge, APACHE4 and 5, which are ADAS SoCs empowering cameras with various “smart” functions. Because the products are designed to be located in a vehicle’s camera module, they have very low power consumption and are very small. The demonstration will show the SoCs’ various use cases, along with the well-organized APACHE5 SDK. (Representative: James Kim, Technical Marketer(Director))

Pick a Card, Any Card: Using ML-Enhanced Vision for ID and Label Tracking

Security cameras can play a valuable role in tracking the whereabouts of goods and personnel within facilities, but to do so they need to be able to reliably read information on ID cards and labels. Using a deck of standard playing cards, Perceive will demonstrate how the combination of a high-definition image sensor and the company’s Ergo edge inference processor can read information at distances beyond the capability of the human eye and expand the role of computer vision in applications such as inventory control and access management. (Representative: Kathy Cook, VP, Business Development)

Smart Video Conferencing on the Edge

Enhancements in AI processing capabilities on edge devices are enabling a richer video conferencing experience. These capabilities are leveraged in use cases that range from biometrics and automatic access of calendar and video conferencing applications through face and voice identification to a smarter framing of the scene in front of the camera, and understanding the language for auto-subtitling...all on the edge. The Synaptics VS680 and VS780 SOCs were designed with smart video conferencing use cases in mind. They are highly integrated to implement cost-optimized video conferencing devices in different form factors and boast the needed AI computation. In this demonstration, the VS680 will run smart framing, where the captured scene in front of the camera is dynamically adjusted to show the region of interest using machine learning. (Representatives: Zafer Diab, Director of Product Marketing & Xin Li, FAE Engineer)

Real-Time Video Post-processing Using Machine Learning

Enhancements in AI processing capabilities on edge devices are enabling significant enhancements in video scaling and post-processing versus, compared to what had been possible with traditional scaling integrated in SoCs. These enhancements enable the scaling and post-processing to be theme-based. Scaling for sports content can be adjusted for the high motion content, for example, while scaling for video conferencing can be optimized for the (mostly) static and moving face content. Synaptics SoCs integrate an internally developed machine learning engine called QDEO.ai, which performs super-resolution scaling that takes the theme of the video into consideration. This demonstration will show a side-by-side comparison of scaling using the traditional hardware scaler and the QDEO.ai scaled video. Quality enhancements are accentuated when performed on lower-bitrate input video, such as with videoconferencing. (Representatives: Zafer Diab, Director of Product Marketing & Xin Li, FAE Engineer)

From Neural Network to Edge with Synaptics

Synaptics will present an overview of the Katana Edge AI processor and tensaicc, the compiler that Eta Compute is developing for use with Katana. After we introduce Katana's novel multicore architecture, you'll see a live demonstration of how tensaicc compiles a neural network and generates power-, cycle- and memory-optimized code that takes advantage of the architecture. (Representative: Vineet Ganju, VP/GM Audio Business Unit)

Simultaneous Localization and Mapping (SLAM) Acceleration on DesignWare ARC EV7x Processors

Simultaneous Localization and Mapping (SLAM) creates and updates a map of an unknown environment while at the same time keeping track of an agent's location within it. SLAM is a key component in a variety of systems, such as autonomous vehicles, robotics, and augmented and virtual reality. This demonstration shows how to accelerate a SLAM software stack by offloading some of the processing to the ARC EV7x processor, as well as how adding an extra core to parallelize the algorithms can further increase the acceleration. We’ll also show how, with an EV7x-based system, you can incorporate a deep neural network engine to expand the system's intelligence. (Representative: Liliya Tazieva, Software Engineer)

SRGAN Super Resolution on DesignWare ARC EV7x Processors

Image super-resolution techniques reconstruct a higher-resolution image from a lower-resolution one. Although this can be done with classical vision algorithms, the results are inferior to what neural-network-based solutions can now offer. In this demo, we show how a Generative Adversarial Network can be used to intelligently infer the missing pixel data and generate a high-quality high-resolution image. The demo will run on the ARC EV7x processor and showcase the Deep Neural Network accelerator engine. (Representative: Liliya Tazieva, Software Engineer)

X-Ray Classification Solution for COVID-19 and Pneumonia Detection

Spline.AI is collaborating with Amazon Web Services and Xilinx to deliver an open-source, open-model, X-Ray classification solution for COVID-19 and pneumonia detection. This demonstration will show the model, deployed on the Xilinx Zynq UltraScale+- MPSoC device-based ZCU104 evaluation kit and leveraging the Xilinx deep learning processor unit (DPU), a soft-IP tensor accelerator that is powerful enough to run a variety of neural networks, including classification and detection of diseases. (Representative: Quenton Hall, AI Architect)

Out-of-the-Box with the Kria KV260 Starter Kit: Up and Running in Under an Hour

This demonstration will provide a detailed look at the Kria KV260 Vision AI Starter Kit and its companion basic accessory pack. We’ll show you just how quickly and easily you can get our Smart Camera accelerated application up and running, with no FPGA experience required. (Representative: Karan Kantharia, Product Line Manager)

Visual Machine Learning and Natural Language Processing Fusion on the Xilinx Kria SOM

This demonstration shows the new  Kria K26 SOM running both vision machine learning (ML) and natural language processing on the same Xilinx Zynq-based platform. This solution showcases the dynamic switching capabilities of the Xilinx deep learning processor unit and integrates audio ML keyword-spotting to control the video display. (Representative: Girish Malipeddi, Director of Video and Imaging Solutions)