AI large models drive the acceleration of cycle evolution, and 3D visual perception opens up space o

Mondo Technology Updated on 2024-01-31

Driven by the AI model, investors and the industry are thinking about the impact and opportunities it brings. Because it gradually has the ability of "intelligent emergence", how the industrial chain side undertakes these technology empowerment and the selection of investment nodes has also become important.

One of the major themes of investors in recent years has been AI. Zhang Chen, general manager of Songling Investment in a village, told the 21st Century Business Herald reporter, "Through the accumulation and learning of AI direction, our company has formed an ecological approach: when perceiving the cycle, we attach importance to the layout of AI vision, touch, smell, and brain-like fields;."Now is the model cycle, and the future will move towards the behavior cycle. Regardless of the stage of development of the AI cycle, we are firmly optimistic about China's industrial development. ”

In addition to ChatGPT, which is biased towards the natural language side, in 2023, Meta will release its image AI large model SAM (Segment Anything Model). According to reports, this is a basic model for image segmentation, which can realize the ability to distinguish objects and understand images without additional training and zero-shot generalization. The industry believes that this has opened the GPT moment of machine vision, which will promote the application of cross-vision scenarios, such as autonomous driving, security monitoring, etc.

In the first half of last year (2023), I still have some concerns and even a sense of crisis: when AI is so powerful that any data input can get good results, is it possible that we will not need the 3D perception industry?Zhu Li, founder and CEO of Guangjian Technology, analyzed to reporters, "But through exchanges with the Silicon Valley industry, our unanimous conclusion is: First, AI is very dependent on data**, if the input data is very poor, it will lead to bad output results, so good sensors are very valuable;Second, AI ultimately needs to be computed and cost should also be taken into account, and it is important to find a balance between sensors and computing power. ”

The emergence of generative AI has shortened the duration of different AI development cycles, and also driven the evolution of the industrial chain to open up new space.

AI-driven evolution

If the AI model is similar to the brain that performs calculations, it needs sufficient tentacles and nutrients to support the operation of the brain, so it is very important to develop and grow the related industrial chain around the AI perception side.

Tianfeng pointed out that it is optimistic that 3D vision is expected to achieve a more efficient intelligent perception and control system through closer integration with AI. By applying AI technologies such as deep learning, machine learning, and large language models to 3D visual perception, more efficient data processing, feature extraction, and pattern recognition can be realized.

Zhang Chen analyzed to reporters that in the three cycles of AI development defined by him, the perception cycle stage of AI is mainly to enrich the relevant information required for embodied intelligence by perceiving the external environment, such as vision, smell, brain, etc., and the core is to solve the problem of information acquisitionThe process of solving machine thinking and decision-making problems in the model cycle;The behavioral cycle may be an application that is currently unattainable by human cognition.

The emergence of the SAM model opens up a new space for the application of the industrial chain based on graphics and vision.

SAM Model Outline).

At present, the development process of the model cycle is shortening, such as the study of multimodal fusion, if the investment institutions do not deploy in the early stage of 2021, there may be no opportunities in the future. Because AI is developing so fast, knowledge needs to be updated almost every week. Zhang Chen lamented that at present, the AI industry has entered the model cycle, and there are different types of basic models and industry models, and there will be many investment opportunities in the application stage of large models.

Under this logic, Yicun Songling faced the perception cycle and began to participate in the investment of Guangjian Technology, a leading company in the field of 3D vision, at an early stage. In the recent 200 million yuan Series B financing announced by Guangjian Technology, CICC Capital, Yicun Songling, Chongqing Kexing and other institutions participated in the investment.

According to reports, Guangjian Technology was established in 2018 and has completed six rounds of financing so far, almost a new financing every year after its establishment, with a total financing amount of more than 500 million yuan from the initial millions of dollars to hundreds of millions of yuan today.

Zhu Li used to be the head of Apple's 3D sensing project, and chose to return to China to start a business after exploring the growth space in this segment. He told reporters that 3D visual sensing means improving the perception ability of the machine and constructing X-Y-Z spatial coordinates. Compared with 2D vision, 3D vision mainly serves machine algorithms, which is more accurate and secure, and also has stronger privacy, which can bring better perception capabilities to artificial intelligence.

The emergence of a large AI model expands the original capabilities of the machine. For example, when designing household sweeping robots before, the obstacle avoidance function usually needs to be implanted with a variety of classification models to have better performance, but with a general large model, the machine can be quickly improved with a model. ”

Therefore, for the visual sensing industry, the emergence of AI large models can drive more complete products, and then promote the large-scale application of AI-related software and hardware. The SAM model mainly solves the problem of the versatility of the perception layer, making the process of transforming spatial sensing information into natural language processing more efficient. "Frankly, it's hard for most application companies to build a big foundation model on their own, but by introducing this underlying capability and embedding our data on some vertical applications for fine-tuning, we can bring differentiated value. He further noted.

Of course, in the application of large models, we are still on the application side, so we don't need to be in a hurry to participate in it early. Zhu Li said that it is possible to become an application participant of the large model and explore new directions, but there is no need to invest too early in the large model itself, such as fine-tuning and other actions. Because it's very likely that what you're doing now will be offset by Open AI's functional upgrades in a few months, it's critical to capture the core of your own vertical scene.

Perception side roadmap

In the face of the surging wave of AI, the technology and application roadmaps of industry chain manufacturers are also gradually expanding.

Zhu Li said that the background of this Series B financing is that the company is at a rapid business growth node. On the one hand, the current business growth needs capital injections to support a larger market share;On the other hand, through a certain amount of capital reserves, the company can build deeper barriers in the three key links of perception, algorithm, and computing.

I'm constantly thinking. Visual perception is almost related to intelligence-related industries, and what to choose to do is first to locate what is created in the value chain. Zhu Li analyzed that because of the current real-life artificial intelligence applications, more than 85% of the information comes from visual information. Guangjian's goal is to solve the problem of interaction between smart devices and people and spaces through visual capabilities.

He continued that Guangjian Technology has built a "first-class library", and the ultimate goal is not to predict which technical route the market will choose, but to reserve capabilities and products first, and then based on the needs of the industry, help the market make a good choice of technology, and guide the market to a more effective way to land, so that technology can truly empower the industry and the market.

The mobile phone is the first application to push the 3D vision industry to an inflection point, and Apple took the lead in applying Face ID in the iPhoneX to reduce the cost of the first chain quickly, so that there is the possibility of further exploring applications in other industries.

In 2023, there is an obvious trend, a number of major domestic mobile phone manufacturers are specially equipped with security chips in flagship mobile phones, in this regard, 3D vision will have greater advantages than 2D vision, and the cost space of flagship mobile phones is relatively high, and 3D vision applications will have more imagination space. Zhu Li analyzed to reporters.

From the magnitude point of view, even if the mobile phone industry is currently in a certain bottleneck development period, it is still a large market with a volume of more than one billion units, for the first chain enterprises, even if it is only used in 10% of mobile phones, it is also the development space of hundreds of millions of mobile phones.

In addition to mobile phones, payment is also a fast-growing market for biometric scenarios. "Face payment is the direction we have invested in the past few years, and palm payment will be the next trend. With the promotion of the industry, European and American countries with relatively cautious information security protection have also recognized this biometric payment method, compared to users may be worried about the privacy of face information, palm payment is considered to be the best form of biometrics in an open society. He continued.

It is reported that at present, Guangjian Technology has reached an in-depth cooperation with WeChat Pay, and has promoted the palm brush technology to transportation, sports, campus, retail, catering, office, shared charging and other scenarios, and the convenience and user experience have been improved compared with face payment.

Application scenarios for palm payment).

In the ups and downs of the XR industry in recent years, Apple's Vision Pro has a lot of built-in optics, which will also be a big opportunity in the field of 3D sensing. But even if Apple will mass-produce its products in 2024, it is clear that this is not yet an application terminal that consumers can accept on a large scale.

Zhu Li believes that the important mission of Vision Pro at the current stage is to provide a large number of professional developers to build an application ecosystem. Because there are no killer applications in the XR industry, the industry inflection point will not come quickly. "Perhaps in 2-3 years, the industry will reach a consensus on the application trend of XR, and at the same time, it is expected that around 2026, it may usher in XR products that the market really needs. Based on this judgment, Guangjian Technology is also currently cooperating with innovative product companies to develop new product solutions, but it will carefully control investment.

Robotics and automobiles are the other two end markets with high growth space and ceilings. According to Zhu Li's analysis, the robot market can bring relatively high added value;"We will currently focus on serving 2-3 domestic customers in the auto market, polishing the products to good enough, and then consider selling products globally." ”

Pathfinding in the cycle

Intelligent vehicles are undoubtedly one of the important driving forces of the current visual sensing industry chain. However, in the face of different scenarios, OEMs may have a process of rapid route selection and adjustment.

Zhu Li analyzed to reporters that there are two types of landing scenarios for 3D visual sensing in smart cars: intelligent driving and human-computer interaction. At present, intelligent driving-related applications, such as assisted driving, automatic parking, etc., are mostly lidar manufacturers, but their high cost means that the threshold for general application is high. Human-computer interaction is also very important in the wave of automotive intelligence, "It is similar to the difference between smart phones and feature phones, whether it is keyboard or screen interaction, the experience is very different." He added that this part of the ability can be extended from the accumulation of the previous consumer electronics field, involving how the car understands the instructions given by people (air interaction, gesture interaction, etc.), how to understand the in-car environment, etc.

For example, when a person enters the car, how the seat automatically adjusts to the needs of the person, we can already deliver this kind of solution. Zhu Li introduced that 3D vision can solve certain privacy and security concerns, such as fatigue monitoring scenarios in the car, which require cameras and algorithms in the car, but 3D vision is not to build image information, but spatial information, "Even if the worst case is captured by hackers, they don't know what kind of image is behind this information." ”

We believe that 3D vision will be a key technology to solve the human-computer interaction in the cockpit in the future, but we will not yet participate in the visual perception of the outside of the vehicle for autonomous driving. Zhu Li added that because of the serious involution of the autonomous driving market, it is important to find the entry point that can realize the commercial closed loop, and it cannot be involuted regardless of the costAt the same time, with the rise of large models, there will be many variables in the future development route of autonomous driving, and the previous architecture design is likely to have major changes, so it is necessary to be cautious at present.

According to reports, in March 2022, Guangjian Technology began to jointly develop cabin 3D vision solutions with domestic new car companies, and delivered a software-based visual perception system before. "In 2023, we have obtained the TS16949 (quality system requirements) qualification, and in 2024, we can provide software and hardware integration solutions. He noted.

Guangjian Technology automotive-grade 3D camera).

From the perspective of industrial development, 3D visual sensing was developed by American and Japanese manufacturers in the early stage and matured, and now there is a relatively scattered situation of industrial chain companies in China.

Zhang Chen analyzed to reporters that allowing AI to understand what the physical world is like is the main advantage of the 3D sensing industry chain, and it is also the key link to provide machines with in-depth information about the physical world. "We feel that machine vision has gone through many rounds of economic cycle changes, and at present, markets such as Europe, the United States and Japan have found the law of development in their segments, and the advantages are obvious. In this direction, from technology to product polishing and mass production, it is necessary to accumulate, iterate, and harvest feedback. The development of machine vision in China, especially 3D vision, also has to go through these processes in order to find the law. ”

The consumer electronics industry is characterized by fast iteration and short cycles, but its explosive power is amazing. Therefore, it is necessary for the entrepreneurial team to have a keen business sense and understand the trend of technological evolution. This is also the reason why we are interested in investing in Guangjian Technology. He said.

It is reported that Guangjian Technology expects to start achieving profitability in the fourth quarter of 2023. "3D vision will be an important bridge for artificial intelligence to enter human life in the future. Zhu Li concluded that the current penetration rate of AI in life is getting higher and higher, and there will be broad room for development in links that require human-computer interaction.

Related Pages