Since the human eye can only see the visible spectrum between 390nm (blue) and 700nm (red) wavelengths, imaging devices can capture a wider range of wavelengths, including X-rays, ultraviolet, infrared, and visible spectrum, with a variety of sensors. Therefore, embedded vision based on imaging sensors has become one of the most popular technologies. Applications include driverless, drone, robot, medical diagnosis, etc. Nowadays, embedded vision is combining machine learning to autonomous intelligent vision. System development, but machine learning needs to be optimized for application scenarios, and traditional development time is long. Now, FPGA leader Xilinx has put a big move, and its product can not only greatly simplify visual-oriented machine learning, but also improve The response time of the visual intelligence system and hardware and software reconfigurability.
1, the reVISION stack turned out, how many cows?
On March 13, Xilinx simultaneously launched the Xilinx reVISIONTM stack, which supports a wider range of embedded software and systems engineers with little or no hardware design expertise to integrate with machine learning, computer vision, and sensors. , so that visually oriented intelligent systems can be developed more quickly.
“reVISION supports the fastest responsive vision system, boosting machine learning inferred unit-power image capture speeds by up to 6x compared to the most competitive computing embedded GPUs and typical SoCs, with computer vision processing The unit power consumption frame rate is increased by 42 times and the latency is reduced to 1/5.†At Xilinx Media Conference, Steve Glaser, Senior Vice President, Strategy and Marketing, Xilinx, stated, “Even without hardware expertise. Developers can also develop embedded vision applications with a single Zynq SoC or MPSoC chip by combining C/C++/OpenCL development processes, industry-standard frameworks, and libraries such as Caffe and OpenCV."
The following diagram is an example of development usability and development time. If a traditional RTL process is used, if Xilinx FPGAs are used, Xilinx provides 20% solution users to solve the remaining 80% of the work, since the game After the software-defined development environment was introduced, the development efficiency was greatly improved. For example, the software-defined database development SDSoc can shorten the development that previously required three weeks to four days. Now, with the Xilinx reVISIONTM stack, machine learning development time can be greatly reduced!
“We have about 1,100 paying users in the SDSoC development environment, and half of them are developing visual applications. We understand that customers want to be able to apply industry-standard libraries and industry application frameworks to help them further improve their productivity, so we launched reVISION Stack, which can greatly reduce development time, using this stack, users can have a high starting point of 80%, they only need to do the remaining 20% ​​of the application development." Steve Glaser stressed, "And, our efficiency is more High, one of the most important indicators in machine learning is image / sec / watt, in this respect Xilinx products are six times higher than other best solutions in the industry, other programs include NVIDIA Tegra, Qualcomm So, etc. In terms of computer vision, my image/second/watt index is 42 times higher than the competition."
In addition, in the field of machine learning, many application performance depends not only on inference but also on response time delay. In this regard, Steve said that Xilinx's reVISION stack solution is only one-fifth of the competitors. In many automotive applications, you can save more lives!
The following figure is an example. If the front car is stopped urgently, if the reVISION stack is used, it only takes 2.7 milliseconds to stop the rear car. And if it is NVIDIA's GPU, it takes 49 milliseconds or even 320 milliseconds to stop. If the car's speed is 65 miles per hour, the distance difference between parking may be from 1.5 meters to 100 meters, and the gap is an accident and no The difference in the accident.
2. How does the customer evaluate the reVISION? stack?
How does Xilinx customers evaluate this revolutionary product that simplifies machine learning development?
“Our Dobby Pocket UAV-AI Edition integrates sophisticated computer vision and machine learning technologies to provide consumers with a unique experience through gesture control and object and subject tracking,†said Yang Jianjun, CEO of Zero Intelligence. “These functions, Until recently, it was only a more expensive system, and we implemented these complex algorithms into Dobby AI with Zynq All Programmable devices. We are very happy to see Xilinx launch the reVISION stack platform, which will make it easier for our team to support our team. Enhance these key computer vision and machine learning algorithms to help us give Dobby AI a more unique personality. Having a deep-rooted technology partner like Xilinx will ensure that we can continue to develop breakthrough solutions in this area."
Shaft Track Roller,Track Rollers,Backup Rollers,Cam Roll
Changzhou Youeryou Trading Co., Ltd. , https://www.farmpartssupplier.com