There was a message from the reader on the same day. Ask us what this means in the end. Is there any big news about mobile AI photography?
Coincidentally, soon the brain has been fortunate enough to be invited to Huawei North Research Institute to visit the new species of mobile photography, and with Huawei's wireless terminal chip product marketing director Zhou Chen and vice president of glory products Xiong Junmin. Communication, got a lot of 'snacks'.
We know that mobile phone AI photography has been hot recently, and many experts believe that the appearance of AI on mobile phones is rewriting the photography rules established by SLR. That is, the goal of mobile phone photography is not necessarily a naked eye, but it is to bring the user a beautiful experience. And the exquisite visual experience. But many friends said that some of the AI photography functions seen today do not have such “amazing” feelings. Why exactly is this?
Although Glory 10 still does not reveal its true colors, through this visit, we can already glimpse the technical features and product value logic of Glory 10. It is proposed that building an AI2.0 model is an important part of this.
From the perspective of the two experts, in order to achieve the true value of AI photography, we must understand three things: Mobile AI must be based on terminal computing power; currently mobile phones under the name of AI have high scores; mobile AI The value must be recognized by all things and derived from the wisdom of all people.
The Mobius Ring on the Chip: The Causal Logic of Kylin Shengkailong
As two representative AI chips, Kirin 970 and Apple A11 are not exactly the same, but they have all been added to the AI task special processing unit. Kirin 970 has added a new calculation in addition to traditional computing units such as CPU and GPU. Unit - embedded neural network processor NPU. These three divisions of labor are clear, just as the GPU is specially launched to perform graphics processing in order to reduce the computational burden of the CPU. NPU is dedicated to the calculation of convolutional neural networks.
Apple's A11 neural bionic chip, built-in neural network engine, to achieve AI calculations run in the terminal.
The underlying technology movements of the two mobile phone giants illustrate that the transfer of AI computing from the cloud to the terminal and the realization of cloud integration are some of the consensus among the giants today. In this way, mobile phones are not limited by the network environment when processing AI tasks. Guarantees real-time and accuracy, but also avoids the risk of data leakage.
In the North Institute's visit, the experts talked about how power is the basis of all AI models today. The lack of computational power, like poverty, severely limits the imagination of developers.
On the other hand, why is this logic of technological value not achieved in Qualcomm until today? Technical problems and strategic planning factors may be one of the reasons. However, the more obvious one may come from differences in business models.
Huawei and Apple are developing mobile phones as their main business lines. Their chip R&D fits their own mobile phone hardware development. Qualcomm, as a global mobile chip supplier, faces the unique needs of the entire mobile phone industry's numerous products. In front of Qualcomm, It is risky to develop an AI special processor, which is likely to add huge costs without the need of a large number of customers. The more conservative way to deal with the changes of mobile AI is, of course, the solution of Xiaolong 845. The calculation of the call angle matches the terminal power. However, as mentioned above, this mode is premised on the sacrifice of precision and power consumption.
Into the paradox of not being AI and not being aggressive with AI, today's Qualcomm and Xiaolong series chips seem to have fallen into a Mobius ring. It is foreseeable that Xiaolong can't be able to get rid of the current product logic for a short period of time, and Huawei is The strategy of customizing mobile chips, which has led to the development of large-scale development of mobile AI, has also begun to emerge on a large scale.
The first, perhaps the so-called mobile AI2.0 model.
Cake or dessert house? Value connotation of mobile AI2.0
In the past few years, the homogenization of the mobile phone industry has become more and more fierce. From dual cameras to curved screens to full screens, mainstream evolution of any mobile phone industry will inevitably lead the industry to follow suit. Both Apple and Huawei have chosen to invest in the AI battlefield. Next, the whole industry began to talk about the story of AI has become inevitable.
After entering this year, the mobile phone has come to an era of 'inorganic AI' competition. Even thousands of Yuan machines have to emphasize that they have AI photography. But what technologies do AI photography come from? What are the actual changes in AI photography? It seems that these mobile phone companies cannot answer.
Of course, a mobile phone equipped with an algorithm, or even an AI filter, cannot be said to be a fake AI. But like the above-mentioned basketball: For AI, true or false is meaningless, the real question is how much can be done. .
In terms of glory, the AI2.0 model should be based on the exclusive AI computing power provided by the Unicorn 970, the imagination of the developers brought by the HiAI architecture, and the camera experience obtained through the long-term machine learning process, and finally be delivered to consumers. The value of AI photography. It integrates many elements such as scene recognition, pattern reasoning and AI effects, etc. It constitutes a truly leap-forward AI experience, and it may be true that the AI2.0 is meaningful to consumers.
This is the biggest difference between the hot AI and AI2.0. One is to stay in a limited number of so-called AI applications, and the other is to provide the possibility to develop countless kinds of AI.
For example, the AI 2.0 experience that AI 2.0 brings to users is like providing a dessert house that continuously produces new products, instead of just providing one or two kinds of desserts that are not exchangeable.
Everything to the People: Taking a Picture of New Species' Reasoning Rhapsody
Then, in the operation mode of Qilin 970's ergonomic blessing and AI2.0 mode, Glory 10 shows us exactly what kind of new AI photography capabilities will not cause great curiosity for the new species of mobile photography. Desired users are disappointed. We might as well open our brains to make a speculative fantasies about this new species of photographing.
According to the news from the North Research Institute's visit, as well as the glorious glory of New West Street in the glory of AI photography in recent 10 years, we can boldly make such conjectures:
The first step is to identify the further enhancement of the photograph. For example, it may produce a new photography experience combined with geographic information. If you visit the Forbidden City, the mobile phone will automatically identify your geographical location, and then tell you the architectural features of the Forbidden City and recommend specific shooting locations and shooting angle.
On the other hand, the current mainstream scene recognition and photographing mode in AI photography is likely to be further strengthened in Glory 10. It may even remove the professional mode, and completely drive the camera with accurate recognition, allowing the white user to completely liberate and bid farewell to the parameters. The era is endless. For example, we take a picture of a landscape with clouds. We hope to be able to have clear colors, clouds and layers, and it is best to have a sense of quality. This is actually true in previous mobile photography. Inaccurate parameters. In glory 10, after the scene is identified, the camera will automatically adjust the parameters and beautify it. We have to do it, just pick up the phone and press the shutter.
In addition, AR camera function is likely to become the new glory of the glory 10. We know that the AR is based on the accurate recognition of the environment, gestures, face and spatial structure, and these recognition capabilities are very important part of the AI algorithm The blessing. In glory 10 recently released 'intelligence', I did not see the AR's ability to take pictures: The subject shot a gesture of concentricity, the photos automatically generate care, this model is estimated to win the girl The crowd is very effective. After taking a picture, click on the color point in the photo to pick it up, and the back of the phone automatically changes to the picked color. It sounds a little magical.
In addition to the AI photography capabilities of the mobile phone itself, Glory 10 may also emphasize the surprise brought by AI Ecology. Many third-party AI applications will surface. This means that users will not only be limited to the AI capabilities provided by the mobile phone itself. , You can also enjoy the wisdom of developers based on NPU and HiAI architecture. Unexpected surprises may follow.
From Kirin 970 to AI2.0 mode, from AI photography to all things AI, maybe we can expect some, Glory 10's AI value experience starts from photography, but it doesn't stop at photography. If you can truly achieve these points, then Called the new species of AI photography, Glory 10 is well-deserved.
Next week, the brain polar body will take a 'special envoy' and unveil the mystery of the new species of glory in 10 mobile phone AI photography. The surprise of this era lurks in the collision between intelligence and all things. Let's look forward to it together. .