Tang Xiaoou passed away, leaving behind a stormy Shang Tang

Author: Ye Hao

Editor: Kang Xiao

Produced by丨Deep Web Tencent News Xiaoman Studio

An AI leader has fallen.

The farewell ceremony and memorial service for Tang Xiaoou, founder of SenseTime, artificial intelligence scientist, director of Pujiang Laboratory, director of Shanghai Artificial Intelligence Laboratory, and professor of Hong Kong Chinese University, was held on the morning of December 19.

Three days ago, SenseTime released an obituary, and the company's founder Tang Xiaoou passed away at 23:45 on December 15, 2023 due to ineffective treatment.

Nowadays, it has become the norm to open doors and pay for faces, and smart security has made some fugitives for many years have nowhere to hide...... In many ways, the dream of AI seems to have come true, and these are more or less related to Tang Xiaoou, who is one of the representative of the many technology promoters who have brought AI technology into daily life.

Tang Xiaoou is known as a leader in China's computer vision and a "pioneer and pathfinder of global facial recognition technology", and the multimedia laboratory of Hong Kong Chinese University, which he single-handedly established, is the "Whampoa Military Academy" of China's AI vision. The initial team of SenseTime came from this laboratory, and the co-founders of SenseTime, Xu Li, Wang Xiaogang, etc.

Many of SenseTime's mid-level employees are students of Tang Xiaoou, and to this day, the company still regularly organizes employees to read papers, which are mainly written by in-house and joint laboratory authors. A middle-level SenseTime told the Deep Web. People who are familiar with Tang Xiaoou told "Deep Web": He is a scientist full of wisdom and humor.

As a relatively out-of-the-circle company in the AI track, SenseTime has been named the "Four Little Dragons of Computer Vision" along with Yuncong, Megvii and YITU. On December 30, 2021, SenseTime landed on the Hong Kong Stock Exchange, and at the beginning of its listing, SenseTime was sought after by the capital market, with a total market value of up to HK $320 billion. It became the number one player in the field of computer vision in China at that time.

In 2023, ChatGPT will be born, and the field of computer vision will not take the lead. An investor in the primary market told the Deep Web: "ChatGPT has rejuvenated the AI track, but the field of computer vision is far from large models in terms of technical framework. ”

In April this year, SenseTime released its own large model, which was approved for launch in August. However, the reality in front of SenseTime is that the commercial landing scene still needs to be broadened. At the end of November this year, Grizzly Bear, a U.S. short-selling institution, released a short-selling report on SenseTime, arguing that SenseTime was inflating its revenue through highly questionable revenue cycles, which caused SenseTime's stock price to hit a new low.

ChatGPT announced the arrival of a new era, and what kind of business story SenseTime will tell under the leadership of Tang Xiaoou came to an abrupt end.

"SenseTime is in an industry that needs to be cultivated for a long time, and at this stage, it is like 800 strong men who stick to the Sixing warehouse. Xu Li once told the author of "Deep Web". At present, it is in a critical period of commercialization of AI companies, and Tang Xiaoou's death has brought more uncertainty to SenseTime, which is encountering commercialization problems.

China's "Leader" in the Field of Computer Vision

Tang Xiaoou is very low-key, and the last time he went public was at the opening ceremony of the World Artificial Intelligence Conference (WAIC) in Shanghai in July this year, Tang Xiaoou used his Tang humor to share the dreams of his three students, Wang Xiaogang, He Kaiming, and Lin Dahua, in the field of deep learning and Xi.

And Tang Xiaoou himself is not a dreamer in the field of computer vision.

Tang Xiaoou was born in Anshan City, Liaoning Province in 1968, with excellent Xi results, with excellent results, he was admitted to the University of Science and Technology of China in 1985, and went to the United States for further study the following year, obtaining a master's degree and a doctorate. In 1992, Tang Xiaoou began to contact face recognition algorithms.

After graduating from MIT with a Ph.D., Tang Xiaoou returned to China to join the Department of Information Engineering at the University of Hong Kong Chinese as a professor, and founded the CUHK Multimedia Lab. It was this laboratory that became Tang Xiaoou's "Whampoa Military Academy". He has profound attainments in the field of computer vision, and is a "leader" in the subdivision of face recognition.

In the decade since the establishment of the laboratory, he is also a Fellow of the IEEE (Institute of Electrical and Electronics Engineers) and serves as the chair of the 2009 Program Committee of the ICCV (IEEE International Conference on Computer Vision).

In 2005, he joined Microsoft Research as the director of the Visual Computing Group, responsible for computer vision-related research. His job requires him to travel back and forth between Beijing and Hong Kong frequently, so it is natural that his 2-year-old son Tang Zhiming will lack care. With so little time with his son, he wanted to record every minute, so he took a lot of photos of his son. As the number of images accumulated to the thousands, he found that finding a moment in his mind became a prominent problem.

Tang Xiaoou sought a solution from his research field that he had studied for many years - using technical means in the field of computer vision to classify and manage photo albums. So he and his colleagues in the Visual Computing Group began to work on a topic called Photo Tagging, using CV technology to classify albums. At that time, CV technology was not yet mature, and Tang Xiaoou started from this simple wish to open the Chinese face recognition technology to practical application.

Three years later, Tang Xiaoou left Microsoft Research and joined the Multimedia Integration Technology Research Laboratory of Shenzhen Institute of Advanced Technology as director and researcher. 

In 2009, Tang Xiaoou and his doctoral student He Kaiming, together with Dr. Sun Jian of Microsoft Research Asia, won the "Best Paper Award" of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) for their paper "Single Image Dehazing Technology Based on Dark Primary Colors", which caused a sensation.

In March 2014, Tang Xiaoou's research team published the GaussianFace face recognition algorithm, which is the world's first computer algorithm that exceeds the ability of human eye recognition, improving the accuracy of face recognition to 99.55%, surpassing the ability of human eye recognition (97.53%) for the first time, and opening the era of technology landing in the entire face recognition industry.

At that time, the top three recognition rates of the LFW (Labled Faces in the Wild) database were all occupied by the three face recognition algorithms of Tang Xiaoou's lab, and Facebook's DeepFace algorithm could only rank fourth. Therefore, in 2016, Tang Xiaoou's Hong Kong Multimedia Lab was selected as one of the world's top ten artificial intelligence pioneer laboratories along with the laboratories of famous universities such as MIT and Stanford, becoming the only team selected in Asia.

The darling of capital has fallen out of favor

SenseTime has caught up with the boom of the capital market since its birth. In the second half of 2014, IDG's Niu Kuiguang threw tens of millions of dollars to find Tang Xiaoou and Xu Li, who were already well-known in the field of computer vision at the Chinese University of Hong Kong, and that year, SenseTime was officially established.

In the middle of that year, Tang Xiaoou and Xu Li, a member of the multimedia laboratory team of the University of Chinese in Hong Kong, founded SenseTime. The SenseTime period before 1600 A.D. was the first dynasty in China to be written, and the founders hoped to use the pioneering spirit of this dynasty to imply the company's original spirit of technology, which is the origin of the name of SenseTime. However, there is also a circulating version that "Teacher Tang" is engaged in business, so it is called Shang Tang.

SenseTime hit the tuyere of capital. In 2017, the hottest thing in the venture capital world was artificial intelligence (AI). According to CB Insights, there were $15.2 billion in global investment in AI in 2017, an increase of 141% over 2016, of which Chinese companies attracted about half of the total investment of $7.3 billion. At this time, AI has become a "sweet spot" in the eyes of investors.

According to statistics, SenseTime has experienced a total of 12 rounds of financing, SenseTime's investors include IDG Capital, Silver Lake, CDH, Morningside, Guangji, Temasek, Tiger, Star VC and other leading institutions, as well as Alibaba, Suning, Qualcomm, Wanda Group and other strategic partners, the last financing before the Pre-IPO is 1 billion US dollars invested by SoftBank, with a cumulative total financing amount of 5.2 billion US dollars (about 33.658 billion yuan)

According to the research report of Southwest Securities, since 2019, the number of investment and financing events in the AI track has declined significantly. A very important reason why SenseTime has hit the capital market since 2019 is the ebb and flow of AI investment in the primary market.

On December 30, 2021, SenseTime Technology landed on the Hong Kong Stock Exchange, and at the beginning of its listing, SenseTime was also sought after, and was once labeled as "revitalizing AI ambitions". The total market capitalization is up to HK$320 billion. However, as soon as the six-month lifting period arrived, some investors and 23.3 billion shares of cornerstone shares before listing were all lifted, accounting for 70% of the company's total share capital.

Specifically, the collapse of SenseTime's market value shattered the technical beliefs established by many investors, who gradually saw the limitations of the current development of AI technology, and in addition, after the listing, SenseTime failed to prove its profitability to the market, and its stock price fell endlessly, which the industry believes is related to SenseTime's profitability.

After SenseTime disclosed the prospectus, it lost 24.3 billion yuan in three and a half years (from 2018 to the first half of 2021) on Weibo's hot search, and now in just one year, SenseTime's loss data has become 40.9 billion.

According to the semi-annual report data, SenseTime's revenue in the first half of 2022 will be 1.415 billion yuan, a year-on-year decrease of 14.3%, which is the first time that SenseTime's revenue has declined since its listing. In terms of profits, SenseTime is still losing money, with losses of 3.43 billion yuan, 4.97 billion yuan, 12.16 billion yuan, and 17.14 billion yuan respectively from 2018 to 2021, and a loss of 3.2 billion yuan in the first half of 2022, with a cumulative loss of 40.9 billion yuan. 

For computer vision companies, technology and data are indispensable, and in the process of obtaining data, image recognition is a very difficult link in the field of artificial intelligence. 

AI technology, which is still in its early stages, has also brought great challenges to the commercialization of AI companies.

In 2020, An Hui, deputy chief engineer of CCID Research Institute of the Ministry of Industry and Information Technology and secretary-general of the Artificial Intelligence Industry Innovation Alliance, announced a data that nearly 90% of the world's artificial intelligence companies are in a state of loss, and more than 90% of enterprises in China's AI industry chain are also in the stage of loss.

At present, the transformation path of AI applications is long. The premise of AI applications is to build real models that are trained repeatedly – writing sophisticated algorithms, collecting massive amounts of data, and spending a lot of time training. As a result, the number of R&D personnel is constantly increasing.

 And this is reflected in the high cost of research and development. According to SenseTime's financial report, one of the reasons for the loss is the high proportion of R&D expenses: according to the prospectus and performance report, SenseTime's R&D expenses accounted for 63.3%, 71.2% and 76.9% of revenue from 2019 to 2021, respectively

 Pan Helin, executive dean of the Digital Economy Research Institute of Zhongnan University of Economics and Law, said that there are strong competitors in most hardware industry segments, with only the advantages of algorithms and AI platforms, but they are similar to "begging for a golden job".

This is also the current dilemma of the entire AI industry. In addition to the large model, now that the highlight period of the entire industry has passed, the sudden departure of the founder has also cast a shadow on the future of SenseTime.