Introduction
Background
The development of computer vision (CV) is an ongoing journey of exploration. Initially, CV focused on basic digital image processing tasks, including pattern recognition, machine learning, and deep learning techniques. In the industry of intelligent video analytics (IVA), there are many application fields requiring common traditional computing algorithms, for example, object recognition, video structuring, and action behavior recognition.
With the development of hardware technologies and algorithms, videos and images have gradually dominated global Internet traffic. With an increasing number of media services, video and image processing based on AI image algorithms becomes a bottleneck in the computing process in terms of cost and performance. In this context, Vision SDK is dedicated to accelerating video and image processing algorithms, improving video and image processing performance, simplifying CV application development, and speeding up CV application development and deployment.
Product Definition
Vision SDK is designed for visual analysis of images and videos, providing basic intelligent video and image analysis capabilities and a programming framework.
- Development through APIs: Native inference APIs and operator acceleration library are provided for you to develop applications. This mode is recommended if you have a fixed application development process so that you can use the algorithm acceleration capability provided by Vision SDK to build CV applications.
- Development through process orchestration: Based on a modular design concept, each functional unit in the service process is encapsulated into an independent plugin. You can quickly build services and develop applications through process orchestration and plugin connection. This mode provides common function plugins, supporting process orchestration capabilities and plugin customization.
Product Benefit
Vision SDK is designed to simplify the development process of inference services for the Ascend AI Processor and lower the threshold for using the Ascend AI Processor.
- Cost reduction and efficiency improvement: Traditional video and image processing is accelerated by NPUs, greatly improving computing performance and reducing costs.
- Easy to use: NPU algorithm acceleration is encapsulated and can be directly invoked by applications, simplifying application development.
Use Guide
This document provides guidance for you to implement functions, such as object recognition and image classification, through Vision SDK API development or process orchestration based on existing models.
You can learn the following information through this user guide:
- Vision SDK architecture, basic concepts, and usage processes of different development modes.
- Methods to use Vision SDK APIs to develop applications and process orchestration plugins to implement applications.
If you are capable of C/C++ and Python languages and have a basic understanding of inference application development, you can better understand the software.
During your initial use, you are advised to see Quick Start to understand the detailed process of each development mode. Then, you can select a development mode that is suitable for your service requirements and learn to develop Vision SDK applications by referring to the corresponding section.