Service packages to offer annotated teacher data for machine learning (News Release)
March 30, 2021
PERSOL RESEARCH & DEVELOPMENT Co., Ltd.
Service packages to offer teacher data for machine learning with annotations that are indispensable to the development of self-driving AI
- Now enabled for full-fledged self-driving vehicle development: everything from vehicle driving to teacher data preparation -
PERSOL RESEARCH & DEVELOPMENT Co., Ltd. (Headquarters: 3-18-1, Sakae, Naka-ku, Nagoya-shi, Aichi Japan, President and Representative Director Eiji Isoda, hereinafter referred to as "PERSOL R&D"), which is involved in the technical engineering business in the PERSOL group, a provider of comprehensive human resource services, has announced that it will start offering service packages from April 2021 to provide teacher data for machine learning1 with annotations2 that are indispensable to the development of artificial intelligence (AI) for self-driving.
1 Machine learning: A technology that learns regularity and judgment criteria from data, and predicts and judges' unknown things based on it
Teacher data: Data about examples and answers in machine learning
2 Such as images and sounds
■Background: Full-fledged development of self-driving cars highlights the challenges of preparing teacher data for machine learning
Advanced Driver-Assistance Systems (ADAS) for automobiles are becoming more sophisticated and the first self-driving Level 3 car model obtained type designation3 in November 2020. Development is in full swing.
Level 3 self-driving cars can automatically drive under specific conditions.4 Self-driving AI accurately performs "cognition", "judgment", and "operation" instead of the driver on the highway. Self-driving AI requires machine learning results for recognition and judgment based on the information collected by sensors that measure and analyze the situation around the vehicle. Preparing teacher data for machine learning, which is indispensable for the development of self-driving AI, requires a huge amount of time and effort. Thus it is a major issue for their development.
Therefore, PERSOL R&D will start comprehensive services. They include everything from vehicle driving on public roads that support not only passenger cars but also heavy-duty trucks, to the acquisition of various data, to the provision of teacher data for machine learning to develop self-driving AI. For these services, we can utilize the know-how and strengths of "ADAS Contract Testing Services" and "Model-Based Development (MBD)"5 that we have cultivated so far.
4 Definition of self-driving cars and government goals https://www.mlit.go.jp/report/press/content/001371533.pdf
5 ADAS Contract Testing Services https://www.persol-rd.co.jp/en/news/notice/20201020-000000.html
Model-Based Development (MBD) https://www.persol-rd.co.jp/en/service/mbd/
■Outline of Our Services
This service collectively provides machine learning data for self-driving AI to automobile developers and suppliers. In response to customer requests, we will set driving routes and carry out actual driving of various vehicles including heavy duty trucks. We measure and acquire data from vehicles, images, and road information during driving. Also, we create "teacher data" which is annotated image data. This service can be customized according to your requests.
■Features of Our Services
- In addition to image data, vehicle driving data is also acquired at the same time
Pedestrian overpasses, tunnels, and bridges on public roads have "height restrictions" and/or "weight restrictions". And you need to acquire specific data for trucks. In addition to image data acquisition, we also respond to requests such as acquiring data on the vehicle and driving environment. Skilled drivers are in charge of driving not only passenger cars but also heavy-duty trucks.
- Annotating your image data is also possible
The amount of image data that can be acquired from the drive recorders is increasing, and it is also possible to annotate the image data that you already have.
- Time saving with annotation automation
This service utilizes MathWorks tools6 to automate annotations so that you can reduce creation time by approximately 60% compared to manual implementation. In addition, it is possible to modify the annotation automation results in more detail by using MathWorks' Ground Truth Labeler.
6 MATLAB® / Automated Driving Toolbox™ / Deep Learning Toolbox™ / Computer Vision Toolbox™ / Image Processing Toolbox™
- Applications other than self-driving
Bounding Boxes and Semantic Segmentation are typical annotation methods. However, in recent years self-driving in things other than automobiles such as ultra-small cars, electric carts, home appliances, etc. has also increased. Therefore, we also support annotations that utilize YOLO v37, which are widely used in their development.
7 AI program that detects the positions and types of objects from the input images
PERSOL R&D will continue to support self-driving development and contribute to safer and more comfortable vehicle development.
■About PERSOL RESEARCH & DEVELOPMENT Co., Ltd. <https://www.persol-rd.co.jp/en/>
PERSOL RESEARCH & DEVELOPMENT Co., Ltd. supports companies in design and development through technical consulting, general and partial contracts, and the dispatch of engineers in the fields of automobile, aerospace, robot, digital consumer electronics, and so on.
And it will continue to offer a "one-stop service" from design and research and development to testing, by using mechanical, electric, electronic and control software design, testing and certification services, and model base development (MBD) technologies.
■About "PERSOL" <https://www.persol-group.co.jp/en/index.html>
Under the Group Vision "Work, and Smile," the PERSOL Group engages in diverse services concerning people and organizations, which include temporary staffing service "Tempstaff," career improvement service "doda," IT-outsourcing, and design development.
We are also working on the next generation of innovations by integrating human resource services and technologies, offering services such as career improvement service "MIIDAS," empowering of technology talent and construction support for DX organizations "TECH PLAY," and a cloud-based mobile POS cashier "POS+."
We will contribute to the realization of a sustainable society and achieving SDGs through business activities in accordance with the Corporate Philosophy and Sustainability Policy.
The fiscal year ended March 2020, the PERSOL Group generated revenue of 970.5 billion yen.
The PERSOL Group was established in 1973 and is headquartered in Tokyo, Japan. PERSOL HOLDINGS Co., Ltd. is the holding company and is listed on the first section of the Tokyo Stock Exchange (security code: 2181).
For inquiries, please click here.