Beijing Daily Client | Reporter Sun Qiru
On July 15, at the 2022 Megvii Technology Open Day held in Beijing, Megvii released its self-developed algorithm production platform AIS (AI Service), and 20 technology DEMOs (technology prototypes) were also on the open day. Appeared on.
Megvii co-founder and CEO Yinqi introduced that Megvii has currently established an AI technology system with two cores: "basic algorithm research" and "large-scale algorithm mass production", and "computational photography" as the core. Algorithm-defined hardware "IoT (Internet of Things) technology system, this AIoT core technology research system will become an important cornerstone to support the commercialization of Megvii AIoT ( artificial intelligence Internet of Things).
When vehicles are driving on the road, pedestrians, vehicles, roadblocks, etc. are recognized in real time, and these calculation processes can be realized with only the power of a power bank. The low-power embedded L2 autonomous driving sensing solution demonstrated at the Technology Open Day attracted many visitors. Technical staff at
said that in-vehicle perception models require high computing power. For example, when performing 3D perception of pedestrians and vehicles under large-resolution input, the amount of calculation is huge. Megvii Research Institute proposed a model super compression algorithm and jointly optimized the algorithm and hardware through software and hardware co-design. The power consumption of the AI computing part of the entire system is kept at about 3 watts. Therefore, only a power bank can be used for the embedded chip to work normally. Successfully implemented a low-power L2 level autonomous driving perception solution.
In addition, 20 technology DEMOs were unveiled that day, including VR bare-hand interaction, hand-drawn character animation, and AI model training that can run with a power bank.
Behind these technologies, the "hands" of R&D personnel are gradually being liberated. On the same day, Megvii officially released its self-developed algorithm production platform AIS (AI Service). Based on the Megvii Brain++ system, AIS has built a zero-code, automated productivity tool platform covering the entire link of algorithm production including data processing, model training, performance analysis and tuning, inference deployment testing, etc., making algorithm mass production truly possible.
It is reported that the AIS algorithm production platform provides a variety of functions to support rapid production deployment of algorithms, which can significantly reduce the threshold for production of algorithms and improve algorithm production efficiency. Currently, the platform can support the training of more than 100 business models, which can be completed in as fast as 2 hours, and the model output accuracy index is much higher than the industry average. It has been verified that algorithm developers using BrAIn++ and the AIS platform can achieve an average acceleration of 30 times for intelligent annotation and 4 to 20 times for automatic learning training. Among the many DEMOs displayed by
at the Technology Open Day, the technology behind "Motion Ape Xiaogang" was developed through the AIS algorithm production platform. This technology can automatically perform high-precision distance measurement, speed measurement and counting when people perform sit-ups, long jumps, rope skipping and other exercises. According to the developers, it mainly uses self-developed models to quickly and accurately detect more than 30 human skeletal points, thereby accurately describing various postures during human movement, thereby completing high-precision ranging, speed measurement and counting. Zhou Erjin, head of algorithm mass production at
Megvii Research Institute, said that algorithm mass production is not a single product, but a conceptual innovation and productivity evolution of the AI production model. It is hoped that through the standardization of AI algorithm production and the construction of an AI productivity platform, the cost and threshold of algorithm production will be greatly reduced, allowing more people to participate and promoting the implementation of algorithms in more industries.
Since last year, more and more companies have proposed AIoT strategies. Based on business practice, Yinqi believes that the implementation of the AIoT industry requires the combination of three elements: First, AI. AI has emerged with deep learning algorithms in the past 10 years. AI is the ability of continuously evolving algorithms. Such algorithms will Various industries and various scenarios build more souls and brains; the second is IoT, if AI is the brain, IoT is the body and the real backbone; the third is space, in the past two or three decades, the virtual economy has been more of a thread Things from the offline are moved online, and the most important thing about the industrial Internet is how to truly define applications and scenarios in the physical space.
