Product Background
The development of computer vision (Computer Vision, hereinafter referred to as "CV") has been a continuous process of exploration. CV originally emerged to enable computers to perform simple processing on digital images. Research in this field mainly covers image processing, pattern recognition, machine learning, and deep learning. In the intelligent video analytics (Intelligent Video Analytics, hereinafter referred to as "IVA") industry, traditional algorithms have many application areas, such as object recognition, video structuring, and action recognition.
As hardware technology and algorithms continue to evolve, video and images have gradually become the main components of global Internet traffic. With the rapid growth of media services, video and image processing based on AI image algorithms has gradually become a cost barrier and a performance bottleneck in computing workflows. Against this backdrop, Vision SDK is committed to accelerating video and image processing algorithms, improving video and image processing performance, reducing the development complexity of CV applications, and speeding up CV application development and deployment.
Product Definition
Vision SDK is an SDK in MindSDK for visual analysis of images and video. It provides basic intelligent analysis capabilities for video and images, as well as a programming framework.
- Develop by using the API. It provides native inference APIs and an operator acceleration library. You can develop applications by calling the APIs. For users with a fixed application development process, you are advised to use this approach and leverage the algorithm acceleration capabilities provided by Vision SDK to build CV applications.
- Develop by using process orchestration. It uses a modular design concept to encapsulate each functional unit in the service process as an independent plugin. You can quickly build services by connecting plugins through process orchestration and develop applications. This approach provides common feature plugins, supports process orchestration, and offers custom plugin development.
Product Value
Vision SDK is committed to simplifying the development process for Ascend chip inference services and lowering the barrier to development on Ascend chips.
- Reduce costs and improve efficiency. It accelerates traditional video and image processing with NPU acceleration, substantially improves computing performance, and reduces costs.
- Extremely easy to use. It packages NPU algorithm acceleration capabilities for direct application calls and simplifies application development.
Usage Guide
Vision SDK user guide helps you implement functions such as object recognition and image classification based on existing models by using either Vision SDK API development method or the process orchestration method.
Through this user guide, you can learn the following:
- Vision SDK software architecture, basic concepts, and the usage process for each development method.
- How to use Vision SDK API for application development and how to implement applications through process orchestration plugins.
If you have C/C++ and Python development skills and some understanding of inference application development, you can better understand the product.
For first-time use, you are advised to first learn the specific process of each development method. See for detailed information about the style and characteristics of each development method, choose the development method that best fits your actual service scenarios and preferred learning style, and then use the corresponding chapters to learn about and develop Vision SDK applications.
Vision SDK software architecture is shown in , and the key modules in the architecture diagram are described in .
Figure 1 Vision SDK software architecture[object Object][object Object]
Table 1 Modules in the architecture diagram [object Object][object Object]
Table 1 Supported product forms
[object Object]