The growth of embedded vision (EV) systems is driving the need for higher performance and energy efficient visual processing capabilities. A number of companies including AMD, CEVA, ImaginaTIon, Intel (Intel), Nvidia and ARM's authorized customers are actively responding to this growing trend, using FPGAs, FPGA/MPU combinations, GPUs and dedicated heterogeneous cores. Different hardware to optimize design tasks.
Synopsys Inc. has released another solution, the DesignWare EV Processor Core (IP) family, designed to be integrated into SoCs with multiple CPUs, whether from ARM, Intel, ImaginaTIon MIPS Other CPUs such as PowerPC are compatible.
The IP core family currently includes two EV52 and EV54 products optimized for visual computing applications, manufactured on a 28nm process. The EV52 is powered by a dual-core RISC processor based on the company's ARC instruction set, operating at frequencies up to 1 GHz, while the EV54 is built on a quad-core to provide higher performance than the EV52. Both products have built-in 2-8 programmable object detection engine processing units (PEs).
Synopsys' vision processor combines an ARC-based RISC core with a convolutional neural network detection engine processing unit.
The EV52 and EV54 processors use the 'Convolutional Neural Network' (CNN) algorithm to inspire the visual information from the human brain to optimize visual computing applications. CNN utilizes a feedforward artificial neural network in which individual neurons are arranged in a manner that reflects overlapping regions within the line of sight. This overlap is an important key for the human eye to track movements, identify environmental changes, distinguish between different objects, and respond to subtle changes in facial expressions.
Mike Thompson, senior product marketing manager at Synopsys DesignWare ARC processor, said: "The EV processor family is designed to perform CNN calculations at 1,000 GOPS/W performance. It only needs to use a fraction of the competitive visual solution. Power consumption enables faster and more accurate detection for a wide range of object applications."
Thompson pointed out that "although there are a variety of visual recognition algorithms competing for favor, we have always considered CNN to have the most significant progress, and we are currently seeing the best object identification solution for target applications, as a camera, wearable device, Ideal for targeted applications such as home automation, DTV, virtual reality, gaming, robotics, digital signage, medical and in-vehicle infotainment systems."
Synopsys is currently working with multiple vendors in the embedded vision market, including Nvidia, CEVA, and Microsoft. However, while CNN is able to achieve more than 95% accurate results, the problem depends on how to achieve this accuracy within the market's acceptable power/performance range.
Thompson said that general-purpose processors (GPPs) can be used for visual processing, but they are too slow due to the lack of advanced mathematical computing resources. Although the graphics processor (GPU) has the necessary mathematical computing resources, it lacks effective mobile vision. The ability of the data to make its visual performance relatively low and the power consumption relatively high.
"Our proposed coprocessor strategy will help reduce CNN costs to an affordable level and the level of power consumption that can be applied to consumer products." Thompson compares a series of typical object detection and gesture applications with For example, the power consumption of an EV processor performing a visual task is approximately five times lower than that of other vision solutions. An SoC with a built-in EV processor consumes approximately 175mW of power when performing a face detection task at a video processing rate of 30 frames per second. In contrast, if the GPU is used to perform the same task, at least 8-10 times higher power consumption is required.
Based on billions of operations per watt, the EV processor (rightmost) performs more efficiently than the object detection and analysis capabilities of other vision solutions.
Embedded Visual Import CNN Algorithm
In order to integrate one or more EV processors into the SoC, Synopsys uses parallel/synchronous operations with the main processor to allow different convolution object detection engines to process through complex and efficient messaging settings and interrupt mechanisms. Communication between the unit and other processor cores (Figure 3). The ARC EV processor can be programmed to automate operations, or when an application must meet specific power/performance constraints, developers can choose to maximize control and function sharing between the EV processor and the host processor.
At the heart of the EV processor is the object detection engine, which contains 2-8 dedicated processing units.
Thompson said: "The number of PEs is configured by the user when constructing the design, just like the streaming interconnect network between PEs - configuring flexible point-to-point interconnects between all PEs. Depends on object detection. The CNN drawing execution on the engine, each point or connection can be dynamically changed."
The architecture is built so that the EV processor memory map can be completely accessed by the host processor, which will allow the host processor to continue to control while allowing all visual processing tasks to be offloaded to the EV unit. Both the EV processor and the EV processor reduce power consumption and accelerate critical visual tasks.
In addition, Thompson emphasizes that this approach also allows a variety of different vision processing units to communicate instantly with the host processor. In order to make communication between EVs and the main processor more efficient, each EV processor can access images stored in the SoC memory mapped area, or through the built-in AMBA AXI standard when needed. The system interface accesses the external memory resources of the chip.
Software suite for CNN development
Thompson pointed out that due to the complexity of CNN, even with EV hardware, it is still a difficult task to launch a suitable combination of algorithms for a specific vision processing application. To help developers mitigate some of their responsibilities, Synopsys provides a complete tool library and reference design that enables developers to more efficiently build, debug, configure, and leverage industry-standard, open source embedded vision toolchain OpenCV. And OpenVX to optimize its embedded vision system.
The optimized tool kit includes an ARC EV processor and more than 2,500 OpenCV features for instant computer vision. In addition, the tool suite offers an OpenVX architecture with 43 standard computer vision cores for edge detection, image pyramiding, and optical flow evaluation, all optimized for execution on EV processors.
Thompson also said that because the EV processor is programmable, it can be trained to support any object detection map and import new OpenVX core definitions. One OpenVX runtime simplifies programming of the processor by allocating the aligned core executions to multiple execution units of the EV processor.
When used to design an EV core, the ARC EV processor can be published and configured via the ARChitect tool. The tool synthesizes RTL that can be integrated into any SoC design to support any host processor, including ARM, Intel, ImaginaTIon MIPS and PowerPC. To further accelerate software development, virtual prototypes will be available for EV processors, as well as support for FPGA-based prototyping, enabling hardware and software co-design before fabricating wafers.
“Embedded Vision is a fast-changing environment,†Thompson said. “Now, CNN seems to be the best direction. But the future is likely to change. In addition to finding solutions that meet the cost and power needs of today's applications. Beyond the program, we want to provide developers with a more efficient way to change their design tasks in the medium term without having to go back to the starting point to redesign."
The HP laptop charger is used to recharge the battery in HP notebook and to provide dependable power for operating the notebook for long periods of time. If the original power charger fails, a replacement AC power charger is necessary, and you can buy all kinds of the HP replacement Laptop Charger from YIDASHUN.
HP/Compaq laptop include hp chromebook series, hp pavilion series, hp envy series, hp eliteBook series, hp stream series and so on. Different specifications of laptop charger Adapter is suitable for different laptop model. Such as 90W 19.5V 4.62A with 4.5*3.0 dc tip, it applies for the HP Touchsmart Sleekbook 15 17 M6 M7 Series; HP Pavilion 11 14 15 17, HP Stream 11 13 14 etc. The common HP laptop charger specification has 30W 19V 1.58A, 65W 18.5V 3.5A, 90W 19V 4.74A etc, and the dc tip has common 5.5*2.5mm, 4.8*1.7mm, 7.4*5.0mm and 4.5*3.0mm etc. Also YIDASHUN can produce new usb type c laptop adapter, there are 20W, 45W and 65W.
Yidahsun's laptop adapter are with smart IC to protect your laptop with over current protection, over load protection, short circuit protection and over heat protection!
All Yidashun hp computer charger is Brand New Replacement Product, works as Genuine parts, 100% OEM Compatible!!
HP Laptop Charger,Compaq Laptop Charger,HP Notebook Charger,HP Computer Charger
Shenzhen Yidashun Technology Co., Ltd. , https://www.ydsadapter.com