Visual vent, foam geometry?

Image source @ Visual China

Text | The Tao always makes sense

In 1969, the United States Bell Labs gave birth to the first CCD image sensor, the industrial vision industry has since set up a boundary of digital images, in this field of long development process, black and white transition to color, resolution from low to high, static upgrade dynamic... The combination of image and vision gradually penetrates into reality into various fields.

The domestic computer vision industry is still prosperous, according to the summary of the six authoritative institutions at home and abroad, the size of the mainland computer vision market is expected to exceed 100 billion; comprehensive conservative forecasts, the number is also about 70 billion, and today, the fourth visual revolution has been upgraded to the concept of 3D.

In March, a large number of 3D vision enterprises have successively won the favor of capital, and Cross-dimensional Intelligence, Landmark Technology, and Yishi Technology have successfully completed large financing, and in the first quarter of 2022 alone, there have been more than 10 financings in the field of 3D vision.

With the drastic progress of the AI field, the entire computer vision industry is shaded by countless outlets, according to relevant data, as early as 2019, its core industry scale and the scale of related industries driven by it were as high as 63.33 billion yuan and 143.86 billion yuan, accounting for 58.2% and 37.6% of the AI core and related industries, respectively.

Perhaps it is the market that is too hot, and the bubbles piled up are slowly overflowing.

"Narrow" scenario?

It is undeniable that the application scenario of computer vision has always been biased.

From the perspective of the proportion of application use, security image analysis is the main application object of the industry, about 68%, followed by advertising and marketing, accounting for about 18%, Internet entertainment and pan-financial fields account for 4% and 7.7% respectively, the remaining mobile phone applications are about 2%, and other innovation areas have accumulated 1%.

The market contradiction caused by the scene restriction is obvious to all, first of all, the head of the enterprise has long been these few business opportunities crazy division, according to forward-looking economic statistics, several major application layer areas of computer vision have hegemons as the king, such as the most fertile security sector, "2018 global video surveillance information service report" shows that Hikvision market share with 37.94% occupies the world's first, Dahua shares ranked second, accounting for 17.02%, Uniview Technology is 2.8% the market share occupies the sixth place in the world.

In addition, Megvii and Yitu in the "Four Little Dragons" are also impressively listed, pan-finance and mobile entertainment are also indispensable to their figures, new players want to re-enter, in addition to digging more application scenarios, it seems that there is no big chance of winning.

It is worth noting that as technology continues to penetrate into life and land on reality, computer vision technology is bound to usher in a wave of scene explosions in the future. According to Tractica's analysis, the top ten most popular use cases of computer vision in 2016-2025, in addition to a few clichéd fields, industry, medicine, autonomous driving and even sentiment analysis... Immediately after the surface.

In particular, industrial manufacturing and the entire driving track dominated by autonomous driving.

The manufacturing industry naturally does not need to repeat, semiconductor electronics, food production, industrial testing, and manufacturing links in various lines all need to rely on visual sensing to gradually achieve automated production. According to the "2021 China Industrial Machine Vision Industry Development White Paper", the scale of China's industrial machine vision market in 2021 is about 25 billion yuan. Looking globally, according to MarketsandMarkets' forecast, the global machine vision market is expected to exceed $13 billion by 2025; it will be close to $14 billion in 2026.

The development process of machine vision in the mainland is quite active, and some institutions have predicted that in the next three years, the compound growth rate of this field will be close to 24%. From the demand side, more than 80% of the current industrial surface defect detection still relies on manual testing, and at least 3.5 million people are engaged in artificial naked eye inspection every day.

Therefore, industry seems to have become the next place of contention for the visual outlet, the reaction of capital has always been the most telling problem, several companies that successfully raised funds in March, the main scope of cross-dimensional intelligence and Yishi Technology and Landmark Technology are in the industrial market.

However, the expansion of the scene is not a temporary effort, why capital poured into the industrial category, is nothing more than because it is impossible to save the narrow application scenario in a short period of time, in contrast, it seems that only the industrial market has material to dig.

As for the driving track, commercial driving safety does provide countless imaginative space for visual applications, it is reported that the domestic commercial vehicle market has continued to expand in recent years, according to the data released by the China Automobile Association, as early as 2018, the sales of commercial vehicles hit a record high, with production and sales reaching 4.28 million and 4.371 million units respectively.

However, it is worth noting that the gap between the application of vision technology in the driving range is not small. It is reported that in the laboratory environment, most companies in the industry can basically identify two-dimensional information such as photos at more than 99%, and the implementation of practical application scenarios is far below this value.

The visual track is crowded with eagerly awaited capital players, but unfortunately, the road ahead is difficult to understand, and the lonely wood is difficult to support.

Helpless "rules of the game"

I have to say that the capital that wants to take the first step in the new round of reshuffle is facing a steady stream of separatist forces in the front and is beginning to be a little impatient. Taking the recent financing case as an example, Landmark Technology, which received tens of millions of financing, was just established in July last year, and Cross-dimensional Intelligence was established in June last year.

New players enter the game, no matter how much merit, the inclusiveness of capital is visible to the naked eye.

In recent years, the visual market is a rare opportunity for enterprises below the head to emerge. Taking the Geling deep pupil, which was once favored by capital, as an example, after 2020, it is the starting point of the "trace" of the gerin deep pupil, and the dual-light temperature measurement intelligent identification device alone hit 47.1602 million yuan in that year, accounting for 19.43% of the revenue of the year.

However, in terms of net profit, the value still can't bear to look at it directly. According to the survey, the company's net profit from 2018 to 2020 was -74.5655 million yuan, -417.5832 million yuan and -78.2016 million yuan, respectively, and the net profit of Geling Deep Pupil in the first half of 2021 was as high as -57.0379 million yuan.

Against the heat and pressure, many companies are in a stalemate, and it is difficult to shake the pattern of the head of the dragon and tiger in the future, and it is impossible to give up this increasingly solid market in the future. At present, the final foothold of the visual outlet is still around the focus of technology, landmark technology, yishi technology and cross-dimensional intelligence This round of financing in addition to team expansion, is product research and development.

The research and development of the technology circle has always been a long-term game that burns money. Relevant data show that the cumulative R&D investment of Geling Deep Pupil from 2018 to 2020 totaled 283.0878 million yuan, accounting for 77.37% of the cumulative operating income in the past three years, especially in 2018 and 2019, the R&D investment far exceeded the operating income.

Loss of money also need to engage in research and development, which is the most helpless game rule in the visual track.

Young enterprises are worried about survival, and there are many troubles in old enterprises. Especially in the world of vision thousands, the security market accounts for half, the traditional security enterprises themselves hardware costs remain high, those rely on security to stand in the field of computer vision most of the enterprises trapped in the profit vortex.

Hikvision occupies a 22.6% market share in the global video surveillance market, and the gross profit margin is the lowest compared with other artificial intelligence companies in the same period, such as iFLYTEK, Megvii Technology, and ArcSoft Technology. Arcsoft Technology, which provides smartphone AI solutions, has a gross profit margin of 94.29%, while Hikvision has less than 50%.

But according to media reports, only 200,000 developers have used Baidu's computer vision capabilities so far.

Coincidentally, Ali's start in computer vision is not late, since 2015, Ali has successively launched hundreds of subdivided products in the fields of image recognition and visual recognition. Nowadays, the Internet, which has always been unprofitable and early, is also a bit confused, and the big visual industry has a difficult scripture to read.

Latecomers are not necessarily losers

In the field of computer vision, from a certain level, the scope of application is more macro, and because of this, the visual war has long been launched in all corners of the world.

First of all, in the entire visual field, the time and speed of overseas entry cannot be denied. Microsoft established an artificial intelligence lab for computer vision technology in the 1990s and published hundreds of papers, and in 2015, Microsoft Research Asia also won three major projects in the ImageNet Large-scale Vision Challenge.

Among them, there are very few mainland enterprises.

Secondly, in the industrial-based machine vision, the current high-end market for global machine vision products is still occupied by American, German and Japanese brands, such as the United States Cognex, National Instruments, The German Basler, Isra Vision, Japan's Keyens, Omron... Among them, Cognex and Keynes monopolize nearly 50% of the global market share.

In contrast, in China, the second half of visual capital finally began to favor the industrial sector, and the warm and slightly blank market background is one of the best explanations. In fact, although the mainland is late in coming, if it is viewed separately, it may not lose.

The more capable mainland is face recognition.

According to the "2018 Global Face Recognition Device Market Research Report" released by the international research agency GenMarket Insight, China will account for 45% of the global facial recognition market share in 2023. The results of the National Institute of Standards and Technology's test of the global face recognition algorithm show that only two of the top 11 companies are from the United States, and the remaining are from China, Russia and Lithuania, of which Chinese companies have won the top five.

There is no doubt that the strong smartphone market and mobile Internet foundation have injected sufficient application possibilities into this field.

Between 2007 and 2017, mainland AI research papers increased by 400%, and at the previous CVPR, Apple and Google sponsored a competition on the image analysis of binocular cameras under different lighting conditions, and the winners came from the National University of Defense Technology of China.

Technology continues to refine, the market is infinitely deepened, but in the visual commercialization, the mainland always lacks a certain field gene. A clear trend is that the overseas visual track has long been mass enterprised, and the mainland is still more academic than technological landing, for example, the number of related papers supported by American companies is seven times that of the official.

It is more gratifying that the mainland's visual technology is gradually moving from "paper talk" into reality and into capital, from 2013 to 2018, in the past 5 years, more than 30 domestic visual field enterprises have been established. This is a good thing, after all, before the torrent of technology upwards, it seems that it is only a matter of time before winning or losing.

Of course, if the bubble had been a little less, the situation would certainly have been clearer.