At present, the scale of the surf technology self-produced database reaches the PB scale, mainly including the automatic driving database and the face recognition database. It is also deployed in the fields of agriculture, industry, and medical care.
In recent years, the AI industry has grown rapidly. The global artificial intelligence market has grown in 2018. Up to 270 billion yuan. Among them, computer vision is one of the most mature application directions of AI, and well-known companies such as contempt, Shang Tang, Yun Cong, and Yi Tu have emerged. Behind all AI applications is massive algorithmic training data. Algorithms such as deep learning used by AI require data for training, and the quality of the data will affect the final output. According to iResearch data, about 20% of the global research and development investment of AI companies is invested in data acquisition.
Founded in 2017, surfing technology is a high-value AI data solution provider. The company found various loopholes in the algorithm by studying various new AI algorithms. Based on years of experience in algorithm and data interaction, the company designed a one-stop solution for data. At the same time, in the data production process, the company uses self-developed algorithms to assist in data processing, such as acquisition, proofreading, labeling, and cleaning, to improve processing efficiency. Finally, through the DaaS platform, the database is matched with the data authorization and the data consumption payment to provide the data service for the client to match the corresponding database. At present, the scale of the surf technology self-produced database reaches the PB scale, mainly including the automatic driving database and the face recognition database. It is also deployed in the fields of agriculture, industry, and medical care.
Surfing CEO Edison tells that global AI companies have great demand for all kinds of high-quality data. When optimizing current product performance and developing new functions, they need to use data for training. With data, it is already difficult to meet the demand. “With the face recognition algorithm we are most familiar with, the face recognition products on the market now have a recognition rate of 99% for the ‘yellow skin, black eyes’ feature of China, but for black people. The recognition accuracy is much lower, because there are not enough black facial data training algorithms,” Edison said. is not only a big demand for data, AI companies also have time-consuming and costly problems when acquiring data. In addition, governments around the world are also tightening their data compliance.One step restricts the use of AI data, which in a disguised way increases the cost of AI’s data acquisition.
Surfing technology is aimed at these pain points and provides AI with high-quality data that can be shared, valued, low-cost, and compliant. The steps of Surfing Technology to produce a database can be roughly divided into:Research Algorithm-Verification Algorithm-Demand Definition-Production Scheme Confirmation-Compliance Confirmation-Resource Acquisition-Data Acquisition- Data Cleaning – Data Annotation – Database Packaging – on the DaaS Platform. In the selection of data collection types, the standard of surfing technology is “forward-looking, mainstream, and universal.” The company has a professional algorithm team to study the current mainstream algorithms on the market, and maintain close contact with customers, pay attention to the future development direction of AI, make pre-judgment, collect relevant data, and make its algorithm as good as possible. % of market demand.
Surfing technology also has different programs and accumulations in its current focus on face recognition databases and the construction of automated driving databases.
In the collection of face recognition database, Surf Technology will choose to select the mainstream 2D and 3D cameras in the market for sample collection, and for the same sample, it will shoot at different angles under different lighting conditions. The face of the different types of mobile phones on the market will be taken to capture the face data of different specifications and different angles. “In general, an ID can generate thousands of images. If there are special requirements from customers, Surfing can perform more detailed shooting requirements. ID pictures can generate tens of thousands of images. Before image acquisition, we also According to the regulations, the collection object is required to sign relevant documents,” Edison said.
At present, Surf Technology has established a global 3D face recognition library, anti-fraud face database, global multi-face face recognition library, cross-age face recognition library, face recognition library under surveillance scene and other databases. The number of products, the number of pictures is on the order of billions.
In the aspect of automatic driving data collection, Surf Technology has adopted a multi-sensor fusion scheme. The visual data of autopilot mainly includes RGB captured by video camera.Image data and point cloud data collected by radar, which require different sensors for acquisition. Surfing technology has obtained the auto-driving data road test qualification. According to the requirements of the collection plan, the company arbitrarily modified the vehicle used to collect data, and also uses the relatively mature mainstream sensor products currently on the market for collection. In terms of processing, the 2D and 3D data joint annotation method is adopted, and the image and the dot matrix data are merged into the same database , and the content includes road condition data of the vehicle during driving, and driving map information. It is reported that the American autopilot company Waymo also recently launched a similar street view database.
At present, Surfing Technology has collected vehicles in both China and the United States, and is building a larger-scale autopilot database with data levels exceeding 10 million frames.
Charging mode, surfing technology is currently in the form of database authorization, the licensing fee is hundreds of thousands, or even millions of yuan, Edison introduced, to the second stage, surfing technology will be officially built and perfected The DaaS platform, which matches the specific size and type of data consumption, provides customers with the best one-stop data service. At that time, different customers can easily intervene in the data of the data set they need.
In the future, Surfing Technology hopes to be a “platform-level” product, receive similar data shared by various companies, reduce the homogenization of different companies, and allow more data to be circulated on its platform. Break the data island.
Customers, Surfing’s customers include Microsoft, Intel, Alibaba, Tencent, Shangtang, and Vision, and other world-renowned companies. In 2019, they earned 50 million yuan, of which about 60% came from overseas.
The founding team of Surfing Technology has worked for Intel, Microsoft, NEC, Nuance and other industry giants. Founder Edison is a continuous entrepreneur who has participated. Stanford Business School Ignite Project.
Before, Surfing Technology’s Angel Wheel Investment of Intime Capital of France is currently seeking 50 million yuan of financing, mainly for business development, product development and operational investment.
————————————————————————————
“Surfing Technology” is a shortlist for financing partner planning technology. If you are an institutional investor and interested in the project, you can add WeChat “Coco@36Kr(day1_36Kr)” to get in touch with the project party. .