When Vertical Scene Falls Into A New Competition Point In The Field Of Computer Vision, "Visual Language Technology" Should Exert Itself In Traffic And Textile Scenes.

2018 is the first year of China's artificial intelligence landing. In the field of computer vision, the most mature technology of AI, Shang Tang, Wan Shi, Yi Tao and Yun have all been listed as unicorns. Their strong financing ability is closely related to their landing applications in security and financial fields.

It can be seen that algorithms and data are not enough to form a competitive barrier. Only in the vertical field can we find the application scene that is combined with technology depth, in order to obtain more room for survival. This is especially important for the new AI venture company.

36 krypton's recent exposure to "visual language technology" is a AI company with layout experience in intelligent pportation and intelligent textile applications. Its core technology is fine grained image recognition, including car pattern recognition, face recognition, commodity identification and defect recognition.

The technology of visual language was established in April 2018. In June of the same year, it received the angel's round financing of Jinsha River venture capital. Its core team members are from the National Key Laboratory of pattern recognition of the Institute of automation, Chinese Academy of Sciences, and has accumulated over 20 years of technology. Past cooperative clients include hundreds of companies and government agencies such as Jingdong, Lenovo, HUAWEI, China Construction Bank, and Beijing Municipal Public Security Bureau.

Founder Wang Jinqiao told krypton that the arrival of the era of artificial intelligence has made the center of technological innovation slowly pfer from research institutes to enterprises, because enterprises have mastered a lot of data resources, machine resources and application scenarios, but enterprises tend to concentrate on product location instead of technology research and development, but AI technology update speed is very fast.

Therefore, the combination of production, learning and research is inevitable.

This is one of the reasons why they came out of the Chinese Academy of Sciences. They can not only take advantage of the original technology, but also integrate into the application scenarios to complete the integration of data, algorithm and industry.

In terms of intelligent pportation, visual language technology has released five kinds of product systems, including vehicle identification and identification system, heavy duty vehicle surveillance and control system, intelligent parking on duty, intelligent toll collection on freeway, and traffic volume analysis.

Specifically, the edge computing smart box developed by the company can be used for face capture in public pportation and hotel, analyzing crowd situation, monitoring and early warning of abnormal driving behavior in public places such as square and station.

Their "intelligent recognition system for heavy trucks" has been put into use in Beijing to monitor the approved truck load.

Wang Jinqiao said that according to the Ministry of pport data, there are about 15 million heavy trucks in China, accounting for 25% of China's major traffic accidents.

Therefore, identifying abnormal conditions such as overweight and fatigue driving is necessary to prevent traffic accidents.

Visual language technology has also introduced a vehicle recognition machine in five provinces in China. It can identify more than 5000 types of vehicle models, search fake and deck cars, and assist the public security departments to quickly locate suspected vehicles, help traffic departments to manage illegal vehicles, and charge large trucks.

Wang Jinqiao told krypton 36 that this product has helped the authorities to catch more than 10 thousand illegal behaviors, including 2800 from wearing seat belts, 980 from driving cell phones, and 220 from smoking.

The car recognition system has also been put into use in Angola and Venezuela.

In addition, the company has won the Wu Wenjun prize for scientific and technological progress and won the "driverless visual perception" track champion in the 2018 AI Challenger global challenge.

Wang Jinqiao said that in the future, they will continue to exert their efforts in intelligent pportation, because the advent of 5G and the maturity of edge computing make it possible to digitalization of traffic, and the foundation of digitalization is big data.

The planning of all the traffic lights in the city and the planning and construction of the intercity main roads require large-scale real-time data. This is the advantage that the former offline survey companies can not get, but the advantage of visual technology.

In the field of intelligent textiles, visual language technology has released an intelligent visual cloth inspection and grading management platform, which can identify 70 kinds of fabric defects. The cloth detection speed is 5 times that of manual inspection, and the recognition accuracy is over 90%.

In contrast, the accuracy of skilled cloth inspecting workers is only 60%-70%, and cloth inspection can also cause damage to workers' vision. At present, many factories are faced with problems such as recruiting difficulties and paying salaries.

Wang Jinqiao said that they now sell a cloth inspection machine for hundreds of thousands of people, and there are 5 textile factories nationwide. The market scale of cloth inspecting machine can reach 60 billion -700 billion yuan.

In addition to the above two application scenarios, visual language technology has also introduced a semi automated annotation platform based on active reinforcement learning, which can replace 95% of human labeling costs.

"If you just give us a product, we will be able to support the identification in second days."

Wang Jinqiao said.

This platform is important because deep learning algorithms rely on large scale precision annotation training data, and most companies have to rely on outsourcing companies to do data collection, selection and annotation, which is a great cost.

In addition, visual language technology has also developed the whole platform neural network forward reasoning engine framework --MNIE. Its automation training, tailoring, quantization and compression tools can optimize and accelerate neural network for specific platforms, and enhance model adaptation capability.

In view of the characteristics of different front-end platforms, the reasoning engine will carry out a large number of specific instructions such as SIMD instruction set and network sparsity at the bottom. It can realize the one key deployment of neural network front-end computing devices for arm, FPGA, x86 CPU or specific neural network accelerator chip (hisi3559a).

The business mode of visual technology is to provide comprehensive visual AI solutions, intelligent products and big data platform for government departments and vertical business customers.

Wang Jinqiao predicts that after 2020, they can break even.

Wang Jinqiao was a research fellow of the National Key Laboratory of pattern recognition of the Institute of automation, Chinese Academy of Sciences, director of the Joint Laboratory of Intelligent Media Computing, director of the Joint Laboratory of big data cloud recognition and director of the Joint Laboratory of video depth recognition.

It is mainly engaged in video analysis and retrieval, large scale target recognition, target detection and tracking, image classification and recognition.

At present, visual language technology has opened the Pre-A round of financing window, and is expected to finance 50 million -6000 million yuan for technology R & D and computing center construction.