Explore the Cutting-Edge AI Software Platform of the Open AI Lab

Table of Contents:

1. Introduction

2. Background and Lab Establishment

• 2.1 Open Air Lab
• 2.2 Harlan Robotics and Sarita
• 2.3 Focus on AI Software for Edge Computing
• 2.4 Headquarters and Divisions

3. AI Software Platform

• 3.1 Hardware Platform
• 3.2 Open-Eye Software Platform
• 3.3 Layers of Open-Eye Platform
• 3.4 Applications and Frameworks
• 3.5 Support for Different Domains
• 3.6 Importance of Dummy Library

4. Optimization of AI Frameworks

• 4.1 Performance Improvement Focus
• 4.2 Dynamic Scheduler for Café
• 4.3 Dynamic Scheduling and Operator Optimization
• 4.4 Debugging and Profiling Tools
• 4.5 Comparison of Results
• 4.6 Introduction to Tianjin Framework
• 4.7 Compatibility with Café and TensorFlow

5. Performance of Tianjin Framework

• 5.1 User-Friendly Interface
• 5.2 Support for Pre-Trained Models
• 5.3 Performance Comparison

6. Domain Library and Algorithm Support

• 6.1 Introduction to Domain Library
• 6.2 List of Supported Algorithms
• 6.3 Upcoming Aligned Library
• 6.4 Development of Server Tools

7. Open-Air Lab Solutions

• 7.1 Open-AI Platform Solutions
• 7.2 Supported Boards and Solutions

8. Resources

• 8.1 GitHub Repository
• 8.2 Contact Information

Introduction

In this article, we will explore the world of open-air labs and the AI software platform developed by a collaboration between Harlan Robotics, Sarita, and Arm. The lab, established in December 2016, focuses on developing AI software for edge computing. With headquarters in Shanghai and a division in Beijing, this collaborative effort aims to provide cutting-edge AI solutions for various devices and computing environments.

Background and Lab Establishment

2.1 Open Air Lab

The open-air lab, also known as "Open I," was initiated by a partnership between Harlan Robotics, Sarita, and Arm. The lab aims to develop AI software solutions for edge computing, focusing on the embodied device and the AI software platform.

2.2 Harlan Robotics and Sarita

Harlan Robotics and Sarita are key players in the development and research of AI technologies. Together with Arm, they have combined their expertise to establish the open-air lab and work on the AI software platform.

2.3 Focus on AI Software for Edge Computing

The main focus of the open-air lab is to develop AI software for edge computing. This includes the software platform for various computing devices, edge gateways, and embodied devices. The lab strives to optimize AI software for inference and is yet to explore age training requirements.

2.4 Headquarters and Divisions

The open-air lab is headquartered in Shanghai, with a division located in Beijing. The Beijing division plays a vital role in the project, focusing on AI research and development.

AI Software Platform

3.1 Hardware Platform

The AI software platform developed by the open-air lab is built on a heterogeneous computing platform. The platform combines CPU, GPU, deep learning accelerators, and DSPs. Various ARM-based CPUs and GPUs are supported, ensuring the platform's compatibility with different computing boards and solutions.

3.2 Open-Eye Software Platform

The open-eye software platform acts as the core of the AI software platform. It consists of three layers: applications, OS, and hardware platform. This platform enables easy development and deployment of AI applications and ensures compatibility across different devices.

3.3 Layers of Open-Eye Platform

The open-eye platform comprises three layers:

1. Applications: This layer consists of application libraries, including computer vision and Speech Recognition. These libraries provide Simplified APIs for developers, enabling them to utilize AI algorithms without deep knowledge of the underlying frameworks.
2. Frameworks: The next layer consists of deep learning frameworks, such as Café and TensorFlow. These frameworks support the development and deployment of AI models by providing a wide range of pre-built algorithms and tools.
3. Computer Library: The lowest layer is the computer library, which includes libraries like OpenBLAS and HCl. These libraries offer optimized functions and operators for different hardware platforms, ensuring efficient execution of AI algorithms.

3.4 Applications and Frameworks

The open-eye software platform supports a variety of AI applications, including face detection, object recognition, and speech recognition. These applications are supported by popular frameworks like Café, Café HRT, and TensorFlow HRT. The platform also introduces its own framework, Tianjin, which is specifically designed for inference tasks.

3.5 Support for Different Domains

The AI software platform aims to support various domains related to sensors and computing devices. Currently, it supports computer vision and speech recognition domains, while more domains are planned for future development. This ensures the platform's flexibility in catering to different application requirements.

3.6 Importance of Dummy Library

The dummy library is a crucial component of the open-eye software platform. It acts as an application library, providing a simplified interface for developers to use AI algorithms without the need to know the underlying deep learning framework. This approach enhances the simplicity and usability of the platform, allowing developers to focus on their applications rather than framework complexities.

(The article continues with headings and subheadings based on the Table of Contents provided.)