In 2024, the popularity of large AI models will continue to increase, and the whole world is looking forward to the changes that large models will bring to us. In order for cutting-edge AI technology to be implemented and for the distant wave of intelligence to enter daily life, some practical industrial channels are needed. Otherwise, AI will end up being a distant seed and it will be difficult to bloom in thousands of industries.
At awe 2024, we can see that manufacturers in various fields are demonstrating AI technology and launching AI products, but there is one work among them that may deserve our more attention. That is the Yueying Vision Solution from HiSilicon.
On March 14, the 2024 China Home Appliances and Consumer Electronics Expo (awe2024) kicked off in Shanghai. With the theme of "Leading New Realms with Innovation", HiSilicon demonstrated its "5+2" intelligent terminal solution to the industry. Among them is "Yueying Vision". There is a judgment in the
ai industry that natural language processing is always popular, but machine vision is what really makes a difference. At the same time, there is also a large amount of data showing that the vast majority of AI reasoning tasks should be completed on the device side in order to maximize the efficiency of AI technology.
The newly upgraded Yueying integrates the latest AI large model technology into the visual chip, integrating keywords such as terminal, cv, and large model into one. What changes will Yueying Vision like
bring to the visual chip industry? How can large AI models bloom in the fertile soil of drones, cameras, and action cameras?
We can walk into awe 2024 together and see the visual feast that HiSilicon has brought to us for the first time on this stage.
Thousands of industries can become intelligent without machines seeing clearly and understanding
With the extensive development of AI technology, machine vision cameras with active recognition capabilities have become standard equipment in fields such as drones and cameras.
However, with the development of industry and technological progress, we will find that visual abilities in related fields still pose considerable challenges. Generally speaking, the vision + AI challenges in the fields of drones, cameras and other fields can be summarized into three points:
1. Implementing machine vision tasks through cloud computing power has brought about the problems of high cost, high delay and poor stability.
2. The CNN algorithm architecture equipped with traditional vision chips has poor generalization and versatility, which has brought about problems such as difficulty in meeting personalized AI needs and excessive AI development costs. The transformer large model architecture with stronger generalization is still difficult to implement in visual scenes.
3. The comprehensive capabilities such as connection capabilities and security capabilities of the vision chip are not ideal, making it difficult to implement scenario-based solutions composed of multiple devices.
These problems have brought many pain points to fields such as security, machine vision, and imaging products. Overall, it is urgent to solidify large AI models on the hardware side and form scene-based visual solutions around intelligent needs.
At this time, Yueying Vision completed an industry-leading upgrade.
When Yueying, the flowers of AI bloom
The new generation of Yueying visual solutions has many evolutionary points in intelligent capabilities. The most eye-catching one is that it solidifies the transformer large model operator on the hardware side, providing hardware acceleration for the transformer to run on the end-side chip, thus achieving a cross-generational combination of large models and end-side vision capabilities.
According to relevant statistics, if based on traditional machine vision algorithms, in a relatively vertical application scenario, users often need to collect more than 1 million data to achieve an accuracy of 98% and a recall rate of 98%. Based on the large open-domain model, it can be fine-tuned with a data volume of about ten pictures for each requirement, achieving an accuracy of 90% and a recall rate of 95%, and reducing the workload of algorithm development by more than 90%. It can be seen from this that Yueying Vision combines large models with end-side machine vision, which can greatly reduce the overall cost of visual AI algorithms, bring more sensitive real-time feedback, and thus comprehensively improve the intelligence level of vision chips. .
large end-side model is one of the four major innovations in the new upgrade of Yueying Vision. Overall, the four major innovation points include:
1. Professional images. Yueying Vision achieves stronger AI computing power and creates a comprehensively upgraded AI ISP capability, allowing the device to achieve better black light full-color effects.
2. Extreme intelligence. The addition of end-side AI large models has greatly reduced AI development costs and improved the application boundaries of machine vision algorithms.
3. Star Flash Internet. is facing the broader short-distance Internet of Things needs in the future. Yueying Vision's solution integrates StarLight technology with faster and more stable connections and AI anti-interference capabilities. With the full rise of StarLight, Yueying Vision will also bring more intelligent value through StarLight and other chip solutions provided by HiSilicon.
4. Full stack security. In terms of the security capabilities of , Yueying Vision’s solution also brings AI technology to the security capabilities, achieving an upgrade in security and privacy capabilities.
Taken together, AI technology has comprehensively contributed to the industrial differentiation of Yueying Vision. At the same time, through comprehensive and multi-level technological innovation, Yueying Vision has achieved industry leadership in AI algorithm application, image quality improvement, and security and privacy. . Technologies in all directions and fields are like flowers, blooming on Yueying and on the road to intelligent upgrading of visual capabilities.
The fruitful
ai technology that allows vision to create scenes constitutes the core differentiation advantage of Yueying. The combination of this advantage and multiple fields finally forms an overall visual solution that adapts to scene-based applications.
At the current stage, the overall upgrade demand for visual chips in many industries, represented by security, comes from three aspects, namely intelligent capabilities, Internet of Things capabilities and security capabilities.
The differentiation of Yueying's comprehensive capabilities also comes from three aspects:
In the field of intelligence , Yueying has implemented chip-level end-side AI technology and completed the solidification of the end-side large model, allowing everything to be recognized and detected. possible.
In the field of IoT and , Yueying has brought Hongmeng native. Therefore, based on Hongmeng's characteristics, it greatly reduces the cost of IoT development of products, and by supporting starlight connection, it brings more powerful short-range communication capabilities in all scenarios.
In terms of security and , Yueying has a new generation of security coding svac3.0, creating a full-stack security solution from chip to coding, system and connection. With the interaction of the three major capabilities of
, Yueying can make intelligent capabilities meet development needs and scenario-based applications, thereby truly promoting the two-way evolution of vision and intelligence in various fields.
HiSilicon coo Zhi Hao believes: "The visual field is where AI first breaks out. HiSilicon integrates end-side AI capabilities and will continue to innovate in the visual field and continue to lead. Yueying Vision is based on SOC, in the original cis, Starlight connection, on the basis of multi-dimensional perception, also truly combines Hongmeng, AI, black light, and full color to form a complete set of solutions."
The combination of these capabilities has allowed Yueying Vision to blossom and bear fruit, becoming a leader in machine vision. I am fully prepared to see the beautiful world.
Let machines understand the beauty of the world
As early as 2018, HiSilicon released the world's first high-definition 8k AI smart IPC chip, thus bringing about an intelligent revolution in the fields of transportation, security and other fields. Today, our lives have deeply benefited from it, and the newly upgraded Yueying Vision, with its more powerful terminal-side AI large models and outstanding comprehensive visual capabilities, will bring great benefits to many fields on the eve of this intelligent era. Come to more profound changes.
In the field of smart management, Yueying Vision’s AI ISP capabilities can make night images as clear as daytime and achieve consistent protection day and night.
is used in fields with strong demand for image visualization, such as drones, robots, etc. Yueying Vision can realize more generalized and universal recognition requirements, so that niche detection targets and recognition tasks can also be realized on the device side in an easy, accurate and low-cost manner.
In the field of live broadcast, Yueying can fully upgrade live broadcast equipment such as action cameras to achieve a live broadcast effect like a SLR, thus giving users a better visual experience. At the same time, Yueying Vision can achieve faster and more accurate focusing, making intelligent preparations for the live broadcast of Momenta Youth.
The Yueying visual solution built with visual SOC as the core can allow more devices and more products to see and understand the world clearly.
And the flowers seen through the eyes of machines will in turn decorate our lives. The beauty of the intelligent world comes from every machine seeing it and every AI understanding it.
