In terms of smartness, domestic Android flagships have more variety, ranging from 2999 to 10999 yuan, each with its own "skills"

Xiaoxiao is from Aofei Temple
Quantum Bit | Public Account QbitAI

Before the Spring Festival, various domestically produced Android flagship phones have been launched one after another, from Xiaomi 12 Pro, Honor Magic V, iQOO 9 Pro, to OnePlus 10 Pro, realme GT2 Pro, moto edge X30... you can't count them all at a glance.

After getting the phone, you can find a lot of interesting functional details:

Xiaomi Mi 12 Pro’s “All Things Focusing” has become a favorite of pet owners . Even if the pet runs out of the picture, it can automatically focus back on the screen. Lei Jun was so excited to repost it after seeing it:

It is like the Honor Magic V that uses the mobile assistant as a "one-card pass". It automatically identifies the location and pops up the application window in special scenarios , such as the delivery code at the express station, the boarding pass at the airport, and the boarding code at the subway entrance:

…

Fancy features don't just appear out of nowhere.

In recent years, competition in the mobile phone industry is no longer based solely on performance parameters and scores. With the development of AI technology and the iteration of hardware such as the Snapdragon 8 chip, AI has been combined with a large number of demand scenarios to provide more abundant possibilities.

So, what kind of AI capabilities do these new features use? And how many of these capabilities are related to the underlying chips?

Let’s take a look at the technology behind these new applications.

From eye tracking to “electronic fences”

Let’s first look at the new applications of AI in photography, office commuting, communications and other scenarios.

In terms of photography , the most interesting feature is Xiaomi 12 Pro's " Everything Focus ".

This technology is a boon for people with poor hand gestures . You no longer have to worry about the target slipping out of the camera because the AI ​​only sees it:

△ Source: Goose Bench

(After discovering that this technology can be used to photograph cats, dogs, fish and birds, netizens began to play around with ping-pong balls and keyboards, and some even wanted to use it to chase and photograph mosquitoes…)

This technology was transplanted from the human vision tracking technology of Xiaomi's "Iron Egg" robot dog and expanded to the Xiaomi 12 Pro.

But it should be noted that the objects tracked by the Xiaomi 12 Pro lens are all objects it has never seen before, and the angles and sizes of these objects will change, and they may even "jump back and forth" in and out of the lens.

That is, AI must master the ability to quickly learn target features, "record" an object when it sees it for the first time, and recognize it when it sees it again.

In addition, Xiaomi Mi 12 Pro's "Everything Focus Tracking" also realizes the " Human Eye Focus Tracking " function for the first time, locking the human eye at close range and zooming in to the whole body at long distances.

△ Even exaggerated movements such as dancing can be recognized (dog head)

When it comes to face recognition, the 300 facial recognition feature points (with a 3x faster speed) provided by Snapdragon 8 can also effectively solve the problem of difficult recognition after the face is reduced under ultra-wide-angle lens.

Combined with the phone's ability to automatically increase shutter speed when it detects intense motion, "Focus on Everything" can maintain the clarity of the image, avoiding effects like the one below:

The improvement in photography capabilities of other flagships is mainly reflected in night scenes.

Mobile phone manufacturers will not miss the advantages that Snapdragon 8 brings in night scene shooting. After all, compared with the previous generation of chips, which only synthesized about 6 frames of images for shooting clear night scenes, this year's Snapdragon 8's AI engine directly synthesizes 30 frames at a time (the AI ​​computing power is strong enough anyway) .

The representative of these is the night scene video of iQOO 9 Pro , which is simply the "king of volume" in night scene mode and a complete bucket of AI algorithms.

From NR noise elimination, MEMC motion compensation, multi-frame fusion, to self-developed AI Deglare and RAWHDR algorithms, there are indeed many AI algorithms that can be used.

If you want to shoot night scene effects faster, you must calculate multiple AI algorithms in real time, which also requires huge chip computing power. Combined with Snapdragon 8, the night scene videos shot by iQOO 9 Pro are closer to human eyes:

For office commuting , the most useful one is the Honor Magic V's smart card .

You will find that this function, which combines the "electronic fence" and the smart assistant, seems to have been running silently in the background. Once you arrive at the designated location, the phone will automatically pop up a notification, fearing that you will miss any important information, but the function does not consume much power.

In other words, as long as you carry this phone with you, when you travel, it is like walking from one "fence" to another, and AI can accurately sense where you have gone.

For example, when you arrive at the airport to catch a flight, an electronic boarding pass or a smart time reminder will pop up on your phone screen to ensure you don’t miss your flight:

This type of function that can be enabled in real time and perceive the surrounding environment, combined with the third-generation Qualcomm sensor hub on the Snapdragon 8, allows AI to process large amounts of data information at extremely low power consumption, increasing performance by 75% while reducing power consumption by 50% compared to the previous generation.

Running this type of real-time activation card on earlier mobile phone chips would not only be slower, but would also consume more power.

In terms of communication , the most practical function is AI network signal enhancement.

moto edge X30 has developed three types of algorithms for signals. ASDiv is used to select the most suitable antenna signal, MAAT is used to increase signal strength in weak network mode (high-speed rail, elevators, etc.) , and finally SRS selects the optimal transmission path for the signal.

Here, Snapdragon 8 integrates the Snapdragon X65 modem and RF system, supports AI antenna tuning technology , uses AI to make signal transmission more stable, while improving hand-grip perception accuracy by 30% and reducing power consumption by 30%.

Well, so playing with your phone in an elevator or on a high-speed train should be smoother than before.

In addition to these scenarios, flagship phones such as Realme GT2 Pro and OnePlus 10 Pro also use AI to schedule resources in games.

Of course, all of the above are still improvements at the application level.

Run more AI algorithms simultaneously

Fundamentally speaking, the improved computing performance and reduced power consumption of these new algorithms are inseparable from the Snapdragon 8's 7th-generation AI engine .

Although manufacturers have launched many new AI algorithms, these algorithms will become useless if the chip performance is insufficient or the operation is stuck.

Therefore, it is necessary to find ways to improve chip AI performance while limiting the size of the hardware.

On the one hand, it is necessary to improve the computing power of dedicated AI chips in hardware. For example, in this year's new generation Hexagon, Qualcomm revealed that the performance of the tensor accelerator has been increased by 2 times, the shared memory has been increased by 2 times, and the overall AI workload performance has been increased by 70% .

On the other hand, further research is needed on software and hardware adaptation. For example, one of the research projects Qualcomm has been conducting in recent years is quantization technology.

Quantization can "slim down" AI models without reducing computing accuracy, allowing them to run on mobile phones with lower power consumption. Currently, many related papers have been published in top conferences such as NeurIPS.

This is also one of the secrets of Snapdragon 8's 7th-generation AI engine to improve performance by 4 times.

The result of improved AI performance is that it helps manufacturers inadvertently cram more AI algorithms into the background.

For example, Xiaomi Mi 12 Pro uses AI to reduce harmful blue light emitted by the screen while keeping the colors natural. In eye protection mode, this algorithm needs to run in the background all the time;

The moto edge X30 is equipped with moto AI ​​Ningyin technology, which is specially used for communication and recording noise reduction , and this will continue to operate in communication scenarios such as making calls and video conferences;

Or the AI ​​self-flow engine and smart charging, which are "imperceptible" AI technologies, all have the support of Snapdragon 8 for real-time computing.

For example, the smart voice assistant, scene perception and other functions in various mobile phones actually require AI computing power to a greater or lesser extent. Snapdragon 8 also plays a significant role in allocating resources and ensuring the smooth operation of algorithms.

However, do you think this is all the AI ​​capabilities of Snapdragon 8?

Nonono.

These capabilities have yet to be “drained”

In fact, many AI functions supported by Snapdragon 8 can still be further explored.

For example, the latest mobile game AI super resolution .

In fact, this has always been one of the technologies that mobile game players are looking forward to.

After all, when mobile games encounter battle scenes or complex scenes that require high frame rates to run, they will basically experience frame drops and reduced image quality to maintain the smoothness of the game, but mobile phone super resolution can solve this problem.

This technology is a bit like the deep learning super sampling technology (DLSS) used in PC games . It uses AI to extract multi-dimensional features of the rendering scene, intelligently combine details in multiple frames, and construct high-quality images.

In this way, when the mobile phone runs the game, it can not only improve the image quality, but also ensure the stability of the game performance, and ultimately improve the gaming experience:

More than just games.

If AI super-resolution can be successfully adjusted to the mobile phone, you can even see clearer live broadcasts , higher-quality special effects photography , and other applications.

In addition to super resolution, the "auto lock screen" function is also worth exploring.

When this feature was released, some netizens joked that it was a " slacking artifact " that could lock the screen in time when the boss was looking at the phone screen.

Specifically, when it detects that a stranger is looking at your phone interface, it will directly "black out" the screen:

I just don’t know if your boss will order you to reopen the interface (dog head) .

This technology is the fourth ISP integrated in Snapdragon 8. With it, it can do much more than just prevent peeping, because it can keep the camera running at extremely low power consumption.

When it detects a QR code , such as an airport itinerary code or a health code for a particular location, it can pop up the corresponding program interface more quickly, saving the time of loading the application.

Well, it can also automatically unlock the interface when you look at the screen——

Yes, the kind that frees your hands.

For example, when you are cooking, even if you cannot free your hands, you can still unlock your phone with your face and view the corresponding recipe app.

Not only that, Snapdragon 8 also supports the simultaneous operation of multiple CV algorithm models, and can also be used to play more interesting special effects photography and other applications.

In addition to the above technologies, Snapdragon 8 also integrates a collection of various tools and models for developing AI functions for its partners.

Including Google Vertex AI NAS function, open source quantitative tools such as AIMET, and thousands of ultra-fast NLP models supported by Baobao Face, all of which are yet to be discovered by manufacturers.

In the future, we may also see these AI features on more new phones.

After looking at these AI applications that have been implemented or may be implemented soon, which one do you like best or what function do you expect to be realized?