Rushing to AI computer vision #Caixin Weekly#-Sino-US

Path: Sino-US >> Cover Story>> 2018 >>
Rushing to AI computer vision #Caixin Weekly#
Almost overnight, the Chinese artificial intelligence (AI) start-ups established by teams led by PHDs in mathematics, computer and statistics have become new unicorns (the term refers to start-ups worth more than $1 billion).  
The start-ups, hot among investors, are mainly focusing on the AI computer vision, a new field in which AI technologies used in computer vision to allow cameras and computers to recognize, chase and measure, such as facial recognition.
In the booming new industry, unicorns compete in the security field, a major demand for the AI computer vision in China, and in the autonomous driving, e-commerce and consumer hardware.
In the 1st issue of 2018, Caixin Weekly, a nationally distributed financial and business weekly, ran a cover story on the AI computer vision, discussing the outlook of the new industry.
Below is an excerpt of the article.
Chinese AI start-up SenseTime Group have sealed investment from US chipmaker Qualcomm and Green Pine Capital Partners in a latest funding round by the end of 2017, which has taken its value to around $2 billion.
“We heard that SenseTime is very popular in the latest funding round, $800 million claimed on the table,” a partner of a famous venture capital, who did not give his name, said.
In March 2014, a team led by Tang Xiaoou, a professor at the Chinese University of Hong Kong, published an article on their new computer algorithm that will improve the machine’s accuracy rate of facial recognition to exceed that of humans. 
Venture capital firm IDG capital grabbed the chance to pour millions of US dollars to support Tang’s team to set up the SenseTime in October that year.
Though the investment aroused little attention that year, three years later, the hot market has turned SenseTime and its peer start-ups, such as Yitu Technology and Beijing-based Face++, into favorites of investors.
Yitu said it raised $380 million in the third funding round in May and Face++ announced it raised $460 million in the third funding round in November. 
Xu Li, CEO of SenseTime, said that one reason for AI computer vision start-ups to succeed was that technology has made the machine’s facial recognition accuracy rate exceed humans, creating a new tool to replace the existing labor force.
In the future, investors would pay more attention to the applications and the future growth space of the AI start-ups, Xu said. “The key of AI is application.”
In recent years, there has been a rapid commercialization of the Internet technology in the subdivided industries, creating rich application areas that have huge demand for AI computer vision.
In China, a special area - the security market, where the government procurement accounts for 60 percent, has brought good business to a number of hardware producers. The market has scaled to hundreds of billions of yuan.
When cameras were installed on a mass scale across the country, public security departments found that the cameras collected lots of data, but the data could not contribute to solve problems, such as sorting out the suspects out of a million people.
The computer vision start-ups seized the chance to enter the vertical industry sector to offer solutions, hoping it will form a closed business cycle of “data - model - application - accumulation of more data.”
In addition to security, potential demand also came from the financial industry, medical care, autonomous driving, e-commerce and consumer hardware. 
In the past two years, Internet giants like Tencent, Alibaba and equipment producers like Huawei also entered the market. Computer vision is considered as a solution technology to be packaged into their projects.
According to a report from Tencent Research Institute in August, there were 146 companies in the computer vision, an AI field with most companies in China, followed by AI robotics and natural language processing.
The computer vision companies raised 14.3 billion yuan, the most, accounting for 23 percent of the total money raised by AI companies.
The computer vision is still at a nascent stage, but too many companies, all focusing on facial recognition, which makes the competition fiercer, said Hua Xiansheng, deputy head of the Alibaba Institute of Data Science & Technologies.
With so many tech companies and in a market where security industry contributes more than 60 percent of demand, what will the future be?
Industry insiders and investors believe that the computer algorithm is the basis, data is the “moat”, capital is the booster, application scenes and the ability to obtain orders decide liquidity. All elements are indispensable.
Security market flourishing again
The first eruptive scene of computer vision in China is security. Almost all computer vision companies were first involved in this field and got their first pot of gold from the government procurement.
The computer vision could offer quite high recognition and accuracy rate in the security field. 
In April, the Ministry of Public Security issued a special plan on technological innovation in the public security field during the 13th Five-year Plan (2016-2020).
Earlier, the central authorities had already issued plans to construct the public security video monitoring system, integrating the cameras and their databases belonging to different government departments and social organizations.
From central to local authorities, they believed that more investment will bring a safer environment in cities, according to a source close to the Ministry of Public Security. 
The source said that Fuzhou, capital city of Fujian province, has recently finished the tendering of 800-million yuan facial recognition project, and Shanghai is expected to invest billions of yuan in public security in next three years.
Yao Song, co-founder and CEO of AI processor startup DeePhi Tech, said that all cutting-edge technologies are usually first adopted in the field with heavy investment - the military industry in the past and the security industry now.
Yao also gave three other reasons for companies to enter the security market: a big enough market, which reached 700 billion yuan in 2017; companies making money quickly from the market existing for many years; and specific segments offering opportunities to get a share.
Tang Wenbin, CTO of Face++, said that at the technical level, the capabilities of the algorithm companies and the security market demand are well matched.
The top application of computer vision technology is the autonomous driving, which has high demand on the machine to finish the process independently, but the security field needs the integration of machine and humans, Tang said.
The machine has clear functions for the security field: “It does not mean the machine will shoot a person directly after identifying a bad person,” Tang said. The machine will find the information and people will analyze it.
The source close to the Ministry of Public Security said that the computer vision is mainly used for catching the fugitives. The facial recognition technology will help police recognize the fugitives by comparing the faces in the database.
The computer vision technology now could create a high accuracy rate of judging the faces in a database of 800,000 photos, and if the database expanded, there would be more errors, the source said.
When companies create more algorithms and models, more application scenes will appear in the security market, Tang said, and the size of the security market will expand several times in the future.
Scenes are critical
The security market gained an explosive growth in 2015, and companies are still looking for areas in which the computer vision technology could be used.
In 2012, Face++ utilized the computer vision technology in a mobile phone game, and though the number of downloads of the game was high on the list of Apple store’s games, Tang said it was not real demand.
“The game could only give users a sense of short-term freshness, but it could not bring long-term value. Similarly, a gesture controlling air conditioning is also false demand,” Tang said.
Lin Chenxi, co-founder of Yitu, said that in 2013, he learned from an official from Suzhou city, Jiangsu province, that the traffic system has a strong demand for recognizing the vehicles. 
The former service provider for the city offered a system which could accurately identify 30 percent of the vehicles, but in practice, that rate has little value and the government wanted a rate of 70 percent.
Yitu offered its solution of a system that can identify vehicles of over 100 brands from photos taken on streets.
The city government decided to allow it to test by identifying five brands in three weeks, and Yitu offered a 95 percent accuracy rate. In three months, the accuracy rate for 100 brands was still 95 percent.
Lin still has the fresh memory of the first order for his company’s system. Later his company got the order on facial recognition system. “Though there were orders, no one believed that the market would boom at that time,” he said.
The financial market demands also came in 2015 when lots of personnel from banking industry started talking about facial recognition that year, Lin said.
Internet banks, such as WeBank, started using facial recognition for users to open an account online.
Xiong Weiming, co-founder of China Growth Capital, said that banks and financial organizations have a huge amount of data, and they strongly need AI to improve efficiency. 
In 2017, almost all banks and many non-banking financial companies used AI to improve efficiency, such as improving the efficiency of credit card delivery and bettering risk control, Xiong said. “The Agricultural Bank of China has promoted the unmanned business halls across the country.”
The AI technologies were widely used in the financial system because bank headquarters promoted the use. The Internet-based operation means the headquarters could centralize powers.
After the financial sector, various specific needs also came from medical care and logistics. 
How to survive
Huang Yan, an investor to SenseTime, said that the Internet giants, Baidu, Alibaba and Tencent, are positively entering the AI market, and if the start-ups could not innovate new business models, they might be phased out.
Computer vision start-ups have few fixed assets except their teams and technology, so why they still raise lots of money in funding rounds? 
An industry insider said that they wanted to obtain more data to form various models of different industries before the competitors joined in.
Forming algorithm models does not need a huge amount of data, but in some fields the data was hard to get, according to industry experts, and the capital and data are still effective “moats” for their companies. 
Shan Shiguang, a researcher with the Institute of Computing Technology under the Chinese Academy of Sciences, said that it was hard to collect data in some barren fields, such as the data on fighting, wandering and stealing.
It was difficult to solve the problems through deep learning, Shan said, and the different demands of different situations have brought pressure to the start-ups in terms of time and labor costs.
He used sorting of wastes in a community as an example to show that training the machine to identify hundreds of kinds of wastes, such as dog shit, vegetable leaves and plastic bottles, would involve a huge amount time. 
The labor cost is also huge for the start-ups. A master’s degree holder would get an annual salary between 300,000 yuan and 500,000 yuan, and a doctor’s degree holder gets annual salary between 300,000 yuan and 500,000 yuan. OR A master’s or doctor’s degree holder would get an annual salary between 300,000 yuan and 500,000 yuan. 
Huang Yan also pointed out that because the business model of these computer vision start-ups is traditional project-targeted model, and the client company might delay the payment to start-ups or turn to other companies.
The Internet giants provide computer vision services to their own businesses first, such as mobile phone Taobao’s function allowing users to search for goods through photos and Tencent’s services in mobile QQ and WeChat. Later, they will offer services to other companies.
Hua said that Alibaba’s computer vision sector offers support to Alibaba’s city cloud brain program that would give cities a new architecture. After producing models and solutions, a platform of computer vision will be established, he said.
Start-ups, such as SenseTime, would like to create their own ecology based on computer vision through standardized products and solutions for vertical industries. They also cooperate with or have invested in hardware suppliers.

  • Prev:No more!
Related Stories
Share this page
Touched Sympathetic Bored Angry Amused Sad Happy No comment

Seeking deeper market penetration, Apple Pay ties up with taxi service in ShanghaiBest Bites of Beijing (Jan 6 - 12)Snapshots of busy people on the moveIs China to blame for the North Korea crisis? The answer depends on who you askWhite snow of Snow Country unable to cover up stain of China's domestic travel industryAccord to bolster global sales of Chinese planesRealty firms enter home rental marketStarting ‘all-encompassing war’ with China topped Trump strategist Bannon’s White House agenda, new book revealsTop-level biosafety lab begins workChina offers 10-year visas to 'high end talent'
< Prev Next >