Standing in the tuyere, can be a pig, but where is the next tuyere?

Absrtact: Standing in the tuyere, can be a pig, but where is the next tuyere? Bill Gates ' answer is: The combination of computer vision (computer vision) and deep learning (deep learning). He came to Beijing in mid-June 2014 and went to the first start-up company

Standing in the tuyere, can be a pig, but where is the next tuyere? Bill Gates ' answer is: The combination of computer vision (computer vision) and deep learning (deep learning). In mid-June 2014 he came to Beijing, the first start-up company is to do computer vision.

This company is called Glitch Deep pupil. This June just won a 30 million dollar A round of financing from Sequoia. Investors are said to have recently tried to invest hundreds of millions of dollars in the B-round.

Let the computer read the world

So far, the image is the largest information carrier of capacity. Thousands of years ago, man invented the word, recorded the voice; 200 years ago, humans made cameras and saved pictures, and until recent decades, humans had invented dynamic images, images, videos.

But it is not easy for humans to fully extract and "read" The information in the image. When you watch a movie, you stare at the male and female characters and notice what color is behind the house? How much information do we miss when we watch TV and play with the phone? When the information of the explosion is stored in the form of images, can humans still be able to collate and retrieve the information?

It's time for the savior to come out.

Computer vision to do very simple, so that the device has "eyes" and "brain", like people to understand the world. Eyes are good, there are cameras, but the machine has no brain, which is troublesome. In the 60 's, researchers began to try to install "brains" into machines. With the computer more and more powerful computing power to do the foundation, can also support a variety of advanced depth learning algorithm.

Computer vision has always been the most active part in the field of artificial intelligence, and fingerprint identification, palmprint recognition and face recognition all belong to the application.

But it has always had two genres. More than 10 years ago, the mainstream view is that with the original two-dimensional camera and powerful algorithms, you can completely restore the real world. Glitch, chief CTO Zhao, and his mentor at Brown University didn't see it that way, "The real world is three-dimensional, how can a two-dimensional camera show the real world?" "Therefore, the two-dimensional camera based on the visual analysis will be greatly compromised."

Zhao's deviant, let him in 10 years later become Glitch Shen Pupil founder. Before returning to business, Zhao has served as senior researcher at Google Research Senior scientist. Google X is the coolest lab in the world, with Google Glass, driverless cars, and the Immortality program, Zhao was one of the first three to propose a Google Glass program.

When a scientist encounters a businessman

Glitch Deep pupil CEO Ho Bo and Zhao is completely two world of people.

April 2013, Blackhawk former general manager Ho Fei quit, back to China to find entrepreneurial partners. At the same time, Zhao is also looking for partners inside Google, two engineers were once very enthusiastic to join, but the final regret.

Two of lost people in the Xiaoping of investors, in Beijing North four ring dilapidated a residential building to meet. Scientists and businessmen chat from afternoon to 2 o'clock in the morning. Scientists have given 80 areas where computer vision can be applied, and businessmen want to try to be the first company to commercialize computer vision.

Initially they thought of the offline retail "electric", so that the traditional retail, such as Ali, Jingdong, such as the owner of user behavior data. But this requires a lot of training data to improve the algorithm, Zhao and how to fly to find the most people, the most images of the field.

Tiananmen Square has the most people, the camera is also the most, the daily production of surveillance video data, carved into a disc stacked up, than the Eiffel Tower is still higher. The data is used for security, and 50% of the world's hard drives are used to store surveillance video. And the security field good money, how to fly to give the data is, last year, China's smartphone sales are 320 billion, and security equipment sales reached 400 billion.

Video surveillance has been the two major problems has been invisible, can not find. With hundreds of cameras in the area, up to only dozens of monitor screen turns to play, monitors often do not see the abnormal screen, after the massive video to find the need for the picture is difficult.

Computer vision can just be solved. Glitch deep pupil through the three-dimensional camera and visual perception technology, you can track the movement of characters, detection action posture, found abnormal active warning, report, security personnel "see". At the same time, you can take the individual action trajectory as a timeline, monitoring, tracking, search, truly "find".

Large market and massive database, it seems that short-term glitch deep pupil can rest easy. However, the days are unpredictable. A key sensor was produced by PrimeSense, which was primesense by Apple in November 2013 and its foreign-mandated technology will expire in 2015.

That night, heard the news of Zhao, he two people are covered. "The first time we think of is to store goods, but also to appease the staff", how to beat fly as memories, "at that time, I began to feel that I and Zhao together." Later, they found new alternative sensors.

Left hand open, right hand closed

Glitch Deep pupil found a Chi market. This October, this set of security products has been laid out, the four major state-owned banks in the first three pilot.

According to the saying, such high-quality computer Vision Technology Limited to the security of the overqualified, so they want to security products based on open platform.

In fact, the security market is also not good to do. Traditional security companies have occupied most of the market share, glitch deep pupil as a start-up company into difficult. Security market and relatively closed, got the Tiananmen Square of all orders, the other square or a talk. In addition, the security industry chain long, from sensors, storage procurement to the integration of components, intruder glitch deep pupil in the supply chain does not accumulate too much.

Glitch's true strength is the recognition algorithm. The best way is to integrate the security identification scheme into a modular solution, open to manufacturers, to do a "closed system, open products", neither with the security vendors to compete, but also to obtain a huge amount of data. Imagine, the future if most of the monitoring screen behind, there are glitch deep pupil of the security identification system, even if people do not know glitch deep pupil This brand, what relationship?

This is the first step. Next glitch deep pupil to do "closed system, open platform", further modular computer identification technology, open API interface. Zhao thought of 80 possible areas of application, but the application of computer vision is likely to have 8,000 or even 80,000, in the way of crowdsourcing open technology, can breed more interesting, meaningful projects and products. Glitch deep pupil to do, only constantly upgrade their own technology, and eventually made an "open system, open platform."

However, it is not so easy to be an open platform. First of all, their own technology to be good enough to abstract out the modularity of things in order to open the API interface. Since its establishment, many people outside the world have questioned the technology of glitch deep pupil. As CEO, he is also aware of this point, frankly, the current technology needs continuous optimization.

The technique of glitch has four directions: limb recognition (including hand movement recognition), face recognition, individual locus recognition, and interpersonal relationship and interaction. At this stage, individual trajectory recognition has been made into a complete product, and has been used in the field of bank security. Hand, limb recognition, is expected to mature after the spring Festival next year. In public places, face recognition does not apply, 1% of the false recognition rate may mean that tens of thousands of people are mistaken for bad, at this stage there is no good solution, is still in the study. The relationship between people and interaction, it is glitch deep pupil in the future most want to solve the problem, people in the online formation of data has been very valuable, that person in the world to form the value of relational data is difficult to imagine.

These four directions are very popular. For example, face recognition, face++ do a little famous. Bat is also trying, but there is no external disclosure details.

Open platform, but also to allow developers to be confident enough. In the domestic, bat do, developers naturally Trust. But a company that was founded two years ago will be hard to make public. This is glitch deep pupil first from the security, do the reason for closed systems. After all, every time Apple releases new products and features, it will find developers in advance to allow consumers to experience these new features in the first time.

It's all around people. Technology mature, glitch deep pupil can also make a car. Car is the extension of human behavior, for example, the driver's mood will affect the driver's driving behavior. On the car identification, glitch Deep pupil has been brewing for a long time, in the just concluded the board, just set down to do this direction, how to operate, but also in the exploration.

Car is the first real sense of the robot, car-related market is also a big market, when people's Travel mode change, people's behavior will change greatly. From people to cars, the imagination is difficult to simply measure and evaluate. In this respect, Zhao's brother, also the founder of Mobileye, has taken a successful step.

In China, the cost of labor is getting higher, and the cost of machines is getting lower. Maybe one day, we will be like Japan, cars cheap, but the taxi will pay a higher cost. When these two cost lines intersect, economic value can be seen even when the business model is not considered.

Next Cisco?

My friends and I were stunned for the first time I heard of investors wanting to give them a hundreds of millions of dollar round of financing. The winter of capital is coming, such voices are heard. Many companies that do technology, few investors have to. The establishment of less than two years of glitch deep pupil, it is possible to step into the 1 billion dollar club. It's no wonder that computer vision is a technical activity. Use the power of capital, gather the best computer vision Daniel, people are the biggest barriers to competition, the latter is difficult to have beyond the possibility of.

However, he flew to tell 36 Krypton that they are not in a hurry to get the money. Entrepreneurship is like a marathon, and rhythm is crucial. Some people are worried about thirst, drinking water early, and may not have good results in the end. Resources may be just as important. Glitch deep pupil already in the physical world of human relations, the future if you can get through the relationship between the virtual world, this one of the imagination space is very big. In the end, glitch to be a great artificial intelligence company.

To become a great AI company, it is necessary to influence as many people as possible in breadth, and to have enough influence and value on individual individuals in depth. Google, Apple, Baidu, Ali, Tencent are such companies. Just now the internet has been hard to come up with such a company. The technological revolution of the Internet is already in the process of completion, the new technological revolution is about to begin.

With regard to the technological revolution, Chris Dixon has the view that the technological revolution typically goes through two main phases: the installation phase and the deployment phase. Each revolution began with a financial bubble that (unreasonably) propelled the rapid "installation" of new technologies; then the bubble burst, followed by recovery, and then, as new technology became more widely "deployed" to other industries and societies, entered a long period of productivity growth. Eventually the revolution was over and the new technological revolution began.

The installation phase is the foundation, the deployment phase is based on the base to do all kinds of practical applications. A large number of entrepreneurial activities emerge during the process of transition to the deployment phase of the installation phase. For example, during the installation phase of the automobile revolution, the activity is to build a car. In the deployment phase, the activities are transferred to the application layer: The expressway system, transportation, urbanization, supermarkets and so on. We are most familiar with the Information technology field, the installation phase we do is the core of information network technology facilities, such as chips, gateways, TCP/IP protocol, the achievements of Cisco, IBM, Intel, Microsoft, the deployment phase is the search (Google), Social (Facebook), electricity (Amazon).

If the next revolution in information technology is artificial intelligence, then now glitch deep pupil is the core of artificial intelligence technology facilities. During the construction of Ober Ipo's core technology facilities, a group of big companies like Cisco emerged. Today, Google, Facebook, Amazon, we are familiar with, Cisco is more a large company image of our memory, but this does not mean that it is not great, it is still a hundreds of millions of dollar market value of the company. Much of what we do today on the internet is thanks to companies like Cisco.

Glitch deep pupil can become the next Cisco, it is really difficult to say, the wind has not risen, the technological revolution is just emerging. Everything is still unknown. The only certainty is perhaps the tide of computer and artificial intelligence. As with the explosion of information, the video field is growing exponentially. Perhaps one day soon, the existing technology will not be able to meet the needs of video search. Companies that deal with video in a traditional way may be as decaying as the portals of the year, and companies that are known for video search will be the new riders.

Perhaps, in the future, I do not search the video, all I need, can automatically appear in front of my screen. and glitch deep pupil, is behind that brain. Since then, the world no longer search, but behind the search everywhere.