SenseTime, The Unsung Hero Behind Various Face Recognition Applications in China
摘要： You might not have heard the name “SenseTime” in the face recognition and artificial intelligence sector, but you’ve definitely have heard of Face VR Special Effect on popular APPs such as Xiaokaxiu, Faceu and Panda TV. It is SenseTime that provides tech support for the function.
You might not have heard the name “SenseTime” in the face recognition and artificial intelligence sector, but you’ve definitely have heard of Face VR Special Effect on popular APPs such as Xiaokaxiu, Faceu and Panda TV. It is SenseTime that provides tech support for the function.
Although SenseTime (in Chinese, Shang Tang, similar to the name of the founding emperor of Shang Dynasty) has a quite Chinese-style name, it owns a world-class team and leading technology in the computing vision field. As early as 2014, SenseTime already defeated Google and Facebook in LFW test, the most authoritative face recognition evaluation system in the world. With a face recognition system whose detection capability even surpassed human eye, SenseTime became a pioneer in the field. In 2015, SenseTime again surpassed Google and won two first prizes at ImageNet, a world-class visual recognition challenge, in terms of “Testing Quantity” and “Testing Accuracy”, becoming the first Chinese enterprise that won first prizes in single indexes.
Although SenseTime has won several awards in international competitions, it is seldom-known back in China. However, it doesn’t mean that SenseTime’s poor in marketing and promoting itself. As a matter of fact, the company is setting a golden example for Chinese startups: one that keeps a low profile, owns cutting-edge technology, has a clear business model and a strong humanistic touch. In fact, SenseTime even has the potential to rise into a unicorn startup.
Key threshold lies in algorithm, not application
In the past, it’s quite common to “copy” each other in China’s innovation or entrepreneurial circle and probe the market with various products of low technological threshold. Once a product works, others would follow suit. Therefore, owing to severe homogenization, the majority of them would be eliminated from the market or merged through capital maneuver.
In an era when there’s lots of fancy ideas and talks but so little cutting-edge technology, it’s natural that the majority of startups would end up in failure after waves of market and capital shift. In this case, we should cherish more those startups that own original technology.
For example, face recognition technology can be divided into two aspects: algorithm and application. However, most people are only familiar with the application aspect. Entertainment products such as Xiaokaxiu and Faceu, as well as safety and some finance products all apply face recognition technology. Speaking of the algorithm aspect, however, few Chinese enterprises are involved in this field.
Owing to the high technological threshold, long-time R&D is necessary in this field. SenseTime is the only Chinese enterprise that focuses on the algorithm aspect and has its own deep learning infrastructure platform. SenseTime’s other counterparts, however, often rely on overseas open-source basic-level algorithm to develop their own applications.
“It’s okay to rely on open-source technological structure in the academic research field, but there are lots of limits as to the business application field. If you rely solely on open-source structures, you would find it hard to adjust some parameters and further polish your products, let alone the fundamental problem: data safety. Based on our 1,200-layer deep neural networking structure platform, our clients can have the same starting point as to Google and Facebook,” Xu Li, CEO of SenseTime Technology, explains.
According to Xu, his confidence in SenseTimes comes from its long-time “technological accumulation”. “It would take even internet giants two years to reach the technological threshold,” Xu told TMTPost.
SenseTime’s strong technological capability, above all, is based on a team of leading talents in the field. As a matter of fact, SenseTime’s team is based on the multi-media lab of Chinese University of Hong Kong and Xu himself was a Ph.D at the lab. At present, SenseTime’s team is composed of graduates from major universities such as MIT, CUHK, Tsinghua University as well as former employees of global tech giants such as Google, Microsoft and Alibaba. Over sixty SenseTime’s employees have Ph.D degrees, the highest in Chinese startups.
“It’s not that we only recruit talents with Ph.D degrees, but that most talents who are, indeed, expert in this field often have Ph.D degrees,” Xu said jokingly.
Another reason why these leading talents are willing to join SenseTime sort of has to do with “patriotism”. As a matter of fact, for a long period of time in the development of computer science, China doesn’t have any initiative over cutting-edge technologies. If global tech giants such as Google continue to take the initiative with the rise of VR/AR technology, then no progress is made by this generation.
Therefore, SenseTime decided to develop its own multi-layer deep learning network at very first. However, the founding team weren’t so sure if they could succeed. After all, it was untapped waters for Chinese startups. However, after they managed to achieve technological breakthrough at last, SenseTime not only set up its own competition barrier over its rivals, but also is able to provide completely different and better service for its clients.
Since SenseTime builds the deep learning network by itself, it has more control over the network and can better satisfy clients’ needs. When providing clients with SaaS services, SenseTime can also grab the data at the backend and carry out more detailed analysis to further improve its service quality.
In industries SenseTime chooses to enter, such as finance, entrance guard, and unmanned driving, etc., clients often have diversified and endless needs over the accuracy of the algorithm, since this has direct relationship with their management efficiency and safety control. Although SenseTime is founded not long ago, its orders have kept pouring in, which also demonstrates the tremendous commercial value of its leading technology.
Getting original technology down-to-earth
Still, as a Business-end service provider, SenseTime might not be well-known at the consumer-end. However, it has accumulated rich experience as to its business model. At present, many Chinese startups with strong technological background would get lost when it comes to business model. In other words, the seemingly monopoly in the relevant industries actually has to do with the lack of mature thinking over an effective business model by most Chinese startup teams. No startup dares to take the lead and rise into a new giant in this burgeoning industry.
Yet, SenseTime’s computing vision technology is not limited to face recognition. Based on its own deep learning network, SenseTime can further expand its business scope.
Although SenseTime’s team are basically Ph.Ds and tech geeks, they also have adequate business savvy.
Based on its own deep learning algorithm platform, SenseTime is able to not only develop productive tools of high efficiency but also help further improve efficiency of the entire industry. According to Xu, artificial intelligence technology itself is like an “artificial brain”, a tool to optimize efficiency of social production. As a matter of fact, SenseTime’s algorithm has surpassed human in various segments. “Our goal is to motivate various industries with our leading technology and drive industrial upgrading through artificial intelligence technology,” Xu told TMTPost.
Up till now, over 200 well-known enterprises have signed contracts with SenseTime, whose major focus at present lies improving user experience and enhancing safety.
As to improving user experience, SenseTime has not only teamed up with smartphone makers such as Huawei and Xiaomi to embed smart photo album function in smartphones, but also with internet platforms such as Sina Weibo, MeituXiuxiu, Faceu and Panda TV to provide AR technological solutions.
Not long ago, SenseTime reached an in-depth cooperation agreement with Xiaokaxiu and would help provide various funny stickers and special effects in live streaming based on its AR technology.
“SenseTime has obvious competitive edge in China as to video recognition and deep learning. We’ve already had in-depth cooperation with SenseTime over face recognition before. Therefore, it’s natural for us to further our cooperation in various other aspects,” a senior executive from Xiaokaxiu explained.
There’s still huge room for imagination and innovation as to combining cutting-edge AI technology with entertainment and internet. It’s crucial to explore the business potential of AI technology and have ordinary users enjoy the benefits of AI technology.
“Democratization of technology has always been a crucial process. Only when new technologies are recognized by the public can relevant research and breakthrough continue. Google’s AplhaGo is a typical example of the democratization of technology, since it gives ordinary people a sense of the intelligence level of artificial intelligence technology. However, technological breakthrough is often based on long-term R&D and technological accumulation. We offer better user experience for C-end clients through our B-end clients. Through SenseTime’s unique business motivation model, our ultimate goal is to help mould the industry ecosystem,” explained Xu.
To achieve this goal, SenseTime adopts a long-term win-win business model. Besides providing solutions, SenseTime charges commission fee based on the number of C-end clients. SenseTime will continue to upgrade its service for B-end clients, so that its partners can continue to offer clients with the best user experience. In return, SenseTime can receive direct feedback from C-end clients to further improve its algorithm.
In other words, SenseTime’s cooperation with partners often lasts for a long period of time. Xu summarized SenseTime’s business model as “B2B2C”: by connecting with C-end clients through B-end partners, we can not only improve our penetration rate more quickly, but also receive client feedbacks faster.
Besides integrating with entertainment and internet, SenseTime’s solutions can also be adopted in safety and personal recognition sector. For example, SenseTime has teamed up with China Mobile and applied its OCR face recognition technology in China Mobile’s user real-name plan; in finance sector, SenseTime’s clients include China Union, JD Finance, China Merchants Bank, etc. and its technology is applied to provide various services such as identity recognition and OCR scanning; in safety sector, SenseTime and Eastday.com established a joint venture to apply cutting-edge technologies such as deep learning in mass analysis products and provide tech assistance as to city safety and management.
Is it likely that a startup like SenseTime will quietly rise into a unicorn one day? Only time can tell. What we know for sure is that the Chinese artificial intelligence market has already been calling for a new giant in the sector.
[The article is published and edited with authorization from the author @Zhang Si, please note source and hyperlink when reproduce.]
Translated by Levin Feng (Senior Translator at PAGE TO PAGE), working for TMTpost.