Wei Jingjing, CEO of Tulin Technology: Four major elements of CV technology commercialization | CCF-GAIR 2017

Leiphone News: From July 7 to 9, 2017, the 2nd CCF-GAIR Global Artificial Intelligence and Robotics Summit was held in Shenzhen, hosted by CCF China Computer Federation and co-organized by Leiphone and The Chinese University of Hong Kong (Shenzhen). In the CV+ special session on the last day of the conference, Wei Jingjing, CEO of Tulin Technology, combined with the development history of Tulin Technology, shared with the guests his experience on how to realize CV (Computer Vision) technology in actual business scenarios.

Wei Jingjing believes that start-ups should proactively explore customer needs, identify customers' real pain points through constant communication and trial and error, and hone their basic skills - algorithms. First, they should start with a single pain point, commercialize the product, and explore new needs on this basis. At the same time, they should accumulate data, create a full set of industry solutions, and establish a solid barrier of business + data + technology.

The following is the full text of Wei Jingjing's speech. Leifeng.com has sorted and edited it without changing the original meaning:

Tulin Technology was founded in 2014. In the three years since its establishment, we have been thinking about the same thing every day, that is, how to make CV technology monetize in actual business scenarios. We have made some attempts and accumulated some experience. Tulin Technology represents the typical characteristics of CV startups - it does not have the high-end resources of SenseTime, nor does it have the strong technical support of Dr. Mei (Leifeng.com Note: Mei Tao, a senior researcher at Microsoft Research Asia). It can only practice and accumulate experience in actual scenarios.

Business scope of Tulin Technology

Tulin Technology's technology covers four aspects:

Face recognition.

According to different needs of users, we divide our services into cloud services, face recognition authorization, security community, etc. At present, products in the fields of Internet finance, banking, and security are being implemented.

Image Identification.

Image recognition must be deeply integrated with actual application scenarios, so there is a relative lack of standardized products in this field. We have come across more than a dozen business opportunities, but based on industry standards, we ultimately chose to land in the two major fields of marketing and insurance.

Image/Video Search.

Video structuring is actually a product that integrates image recognition, video analysis and video retrieval technologies. It can perform detailed analysis and real-time processing of people, vehicles and objects in the video.

The above picture shows our structured analysis of the behavior of cars and people in specific scenarios, such as whether they are riding motorcycles, which is conducive to later information retrieval.

Machine vision.

In the field of industrial machine vision, we combine visual algorithms with customers' actual pain points to provide them with a complete set of solutions. At present, we have implemented solutions such as glass appearance defect monitoring. Nowadays, everyone has a mobile phone in their hands. Traditional mobile phone glass inspection requires human eyes to check one by one. The error of these glasses is often only a few millimeters or more than ten millimeters, which is very boring and painful to check. We saw the workers on the product line holding the screen and checking it bit by bit, so we thought of combining machine vision technology with the pain points of glass inspection to make a detection device. This device has now been put into use and can detect various defects such as white spots and black spots on the glass.

At present, our technology has been applied in practical scenarios including security, Internet finance, banking, securities, marketing, industrial manufacturing, Internet education, and e-commerce, etc.

Four major elements of computer vision industrialization

The industrialization of computer vision needs to focus on four elements: algorithms, data, scene implementation, and solving industry pain points. For small startups, the most direct and effective way to open up the market is to solve customer problems. Computer vision is an industry with high technical barriers, which has both advantages and disadvantages. The disadvantage is that applying technology to scenes is a very complicated process. Therefore, if you just tell customers what problems your technology can solve, it is difficult for customers to understand and it is not meaningful. Only when you make a series of solutions and solve the pain points for them will they pay for it.

Exploring industry pain points

The pain points in the industry are mainly the following:

1. Improve efficiency by optimizing existing business processes;

Second, replace manpower. For example, in the field of security, computer vision can be used to solve problems that previously required manpower.

3. Simplify complex issues;

Fourth, it can be standardized, which is very critical for product design and positioning;

5. The market space must be large enough. Startup companies cannot just solve the pain points of a single customer or a certain type of customer. They must have a type of product to support the subsequent development of the product.

Studying Algorithms

Regarding algorithms, Tulin Technology's initial entry point was image and video search. But computer vision is a combination of multiple algorithms, including image retrieval, optics, object detection, motion tracking, embedded, and so on. Only by integrating multiple algorithms together can we truly solve the pain points of the industry. These are the basic skills we have been practicing in the past three years. Only after the basic skills are in place can we perfectly meet the needs of customers. Only by starting with a single point of technology and accumulating slowly can a startup develop in a larger and more diverse direction and form a complete set of solutions.

Discuss trial and error

Traditional customers usually don't understand what problems computer vision can solve, so we need to discuss trial and error with them. This is an iterative process. First, you need to take the initiative to discuss with customers. If you go directly to the customer's door and tell him what kind of technology you can provide him, he may not be interested. Therefore, you need to understand the customer's business logic, know what your technology can do for him, and guide him to discover deeper pain points. This process is painful, but only by taking a solid approach can you go better in the future. After determining the customer's real pain points, it is also necessary to conduct actual technical research, integrate algorithms with product design, sort out existing business processes, and ensure that there are no problems in every link.

Determine the product form

Existing product forms include cloud, SDK, etc. Whether you want to make a real product or build a platform is something that needs to be considered. The product positioning itself must be good, and commercial issues must also be considered, such as whether customers are willing to accept it.

Perfecting the business model

How can you get a stable income after providing reliable products to customers? An industry will continue to generate new demands during its development, and you need to add these demands to the industry solution. Take Tulin Technology's experience in the insurance and marketing fields as an example. At the beginning, it only cuts into one of the customer's pain points. After commercialization, customers will discuss the pain points of other businesses with you, so we can add it to the existing system.

Establishing business barriers

The biggest problem for most startups, especially computer vision companies, is the lack of data accumulation. How can we get the data and achieve the expected results? This process can be divided into several steps: First, you need to slowly get in touch with the customer, make him feel that your technology is useful to him, and win the customer's trust, so that he will be willing to take out the data. After getting the customer's data, it is necessary to integrate and improve it with the business part to form a product for sale, and actively accumulate data. After the product is commercialized, the data will form a virtuous cycle of accumulation during operation. Only when data and business models form a perfect closed loop can a solid barrier be formed. The most solid barrier is not the barrier of products or business models, but the barrier of business + data + technology.

CCF-GAIR 2017 has come to a successful conclusion. Click to read the original article and view the full report of the conference.