Last year, we covered the impressive AI-powered capabilities of the Mate 10 Pro. This year, the company is starting to really take advantage of its dedicated NPU (Neural Processing Unit).

Before we dive into the nitty-gritty of the features, it’s worth taking a second to appreciate the effort Huawei puts into developing its technologies. In fact, Huawei now spends €10.4 billion each year on R&D to rank it sixth place globally. For perspective, this is even ahead of Apple.

It’s clear the company’s R&D is resulting in real-world performance. Compared against the AI processing capabilities of market-leading chips from Qualcomm and Apple, the Kirin processor developed by Huawei beats them across a range of benchmarks.

We all know benchmarks are one thing, but it’s what the performance offers to end users which matters.

Last year, Huawei introduced what it called the ‘First Generation AI Camera’ in the Mate 10. We were impressed with the speed in which the AI could detect scenes and objects to optimise the camera for the best shots, but the number it could recognise was limited.

Huawei is building on last year’s functionality to offer what it’s calling ‘Pro Skills with Master AI’ to its users. The company aims to offer the capabilities of a dedicated camera with the simplicity expected from a smartphone.

As of writing, it certainly appears to have succeeded. The P20 now tops the DxOMark rankings which are often seen as the closest to an industry standard. This feat was achieved through a combination of innovative hardware and, of course, artificial intelligence.

The onboard AI can now recognise 500+ scenarios across 19 categories. While impressive, you’d be right to say the scene detection is an expected iterative upgrade over last year’s innovation. However, that’s not all Huawei had for us today.

Electronic image stabilisation (EIS) was the first of its kind to come to smartphones. Later, manufacturers took inspiration from high-end dedicated cameras and began to use optical image stabilisation (OIS) which suspends the lens itself to absorb sudden movements. Huawei is now looking to make the next leap with, you probably guessed it, AI image stabilisation (AIS).

The results are pretty spectacular. The AI is really a (serious) advancement of EIS, with the NPU using machine learning algorithms to predict and counteract shaky movements on a frame-by-frame basis.

DxOMark is once again impressed. Ranking the video recording side of the P20 Pro alone, the company gave a score of 98 to set yet another record.

Huawei’s AIS is also used for still photographs to an impressive degree. Anyone with some knowledge of photography will know that in order to get decent night shots you will often find yourself needing to hold a camera very still to allow enough light into the sensor. In the past, especially with smartphones, this required a tripod. Now Huawei claims its AIS is eliminating the need for most scenarios.

A photo taken by Huawei CEO Richard Yu was shown to the audience. At first, it seemed disappointing compared to other shots we’d seen throughout his presentation. Yu then revealed the buildings in the background of his photo were unable to be seen by his own eye yet clearly visible in the photo. Along with the AIS, this would have been aided by the camera’s maximum ISO sensitivity of 102400 (the same as a Canon 5D Mark IV DSLR.) For comparison, the iPhone X’s maximum is 6400.

Motion photography can be notoriously difficult and it often leaves users with blurry photos during the most critical moments. A feature Huawei is calling ‘4D Predictive Focus’ uses AI to keep track of objects in motion and where they’re heading to predict where it needs to focus next for the best shots. We had a brief chance to test out the feature on a person on a swing and the shot was in full focus in contrast to even our Mate 10 Pro despite that device also having a well-respected camera setup.

While this sums up the significant advancements Huawei has made in photography through AI, it’s not all it’s been doing in the field. We had a chance to sit down and interview the company’s director of AI product management and we’ll be posting our insights from that conversation in the coming days.

Are you impressed with Huawei’s use of AI for photography?

The post Huawei uses AI to ‘Master Photography’ appeared first on AI News.

AI-based tasks on most smartphones today are often carried out in the cloud without the average user even being aware. Google Photos, for example, uses deep learning to artistically crop and edit landscape panorama photos like a human photographer. Photos are categorised automatically, and — as of an update earlier this year — the app will even prompt who you may wish to share the pictures with.

Dedicated AI chips will allow faster processing of various tasks on the hardware itself without a need for the cloud.

Current Mobile AI Chips

While still in their infancy, we’re starting to see dedicated mobile AI in handsets before we even head into next year.

Huawei is leading the pack with its Kirin 970 chip which features a dedicated NPU (Neural Processing Unit). The company is claiming its NPU is able to perform an image recognition test on 2,000 images per second, 20 times faster than what the competition can do with a CPU alone.

It’s well known in the mobile industry that new technologies often don’t reach mainstream adoption until Apple gets on board. Fortunately, Apple isn’t far behind and will have dedicated AI cores in the iPhone X’s new A11 Bionic chip to power a range of new features including facial recognition and real-time analysis of 4K video for improving video recordings.

Huawei’s marketing team have been poking some fun at Apple’s expense. After the onstage failure of FaceID at the iPhone X’s announcement, Huawei posted an image of a clown with the caption “Let’s face it, facial recognition isn’t for everyone” alongside the hashtag #TheRealAIPhone ahead of the expected announcement of the Mate 10 Pro on October 16th.

We look forward to seeing whether Huawei can prove itself to be better at mobile AI than Apple, or whether Huawei will ultimately end up looking the clown.

Future Mobile AI Chips

Another given in the mobile industry is whatever Apple has, Samsung will either have or won’t be far behind.

Industry sources familiar with Samsung’s chip plans report the company has been carrying out R&D work into mobile AI chips.

“[Samsung] is in the middle of developing several types of chips that will be capable of processing massive data from AI applications on devices, eliminating the need to communicate with cloud servers,” a source from one of Samsung’s partners told The Korea Herald.

Samsung recently overtook Intel as the leading chip maker which makes this news particularly notable. We’re sure to see a dedicated AI chip debut in an upcoming Galaxy smartphone but we can also expect them in the company’s popular TVs and smart home technology.

“In the next three years or so, smartphones will have designated chips that will improve the processing of AI features by 50 percent from today,” the industry source explained. “A kind of an accelerator can be applied to empower AI devices.”

Investor Excitement

The potential for AI has excited investors who are putting their cash into chip manufacturers set to gain from this new venture.

Nvidia, who relies on the gaming market, is moving into AI and now provides chipsets for some of the world’s largest companies to power their efforts. The gaming market is often cyclical in nature but AI looks set to provide a healthy stream of revenue over the coming years.

The company’s main source of revenue is still the gaming market with $1.03 billion in sales in the first quarter of its fiscal year — although this was down from its average forecasts of $1.13 billion. However, Nvidia reported more than $400 million in quarterly revenue for its data center business selling chipsets to the many companies increasing their computing power for servers with deep learning technology.

These financial results arrive in tandem with a better-than-expected performance from Nvidia’s segment focusing on AI for self-driving cars and other automotive efforts. Combined, this more than made up for the shortfall in gaming revenue and showed investors how AI will provide a huge boon to chipmakers like Nvidia.

As a result, of any company in S&P 500 Index, Nvidia had the largest stock increase in 2016 and its shares have gained 227 percent over the past year.

“AI is the most important technology development of our time, with the greatest potential to help society,” says Nvidia Corp. Chief Executive Jensen Huang. “As the world’s leading cloud providers deploy the world’s best AI platform, with Volta GPUs and Nvidia software, we’ll see amazing breakthroughs in medicine, autonomous transportation, precision manufacturing and much more.”

Companies including Alibaba, Baidu, and Tencent are upgrading their data centers with Nvidia’s Volta-based platforms. It’s clear to us, the invasion of the dedicated mobile AI chips is nigh.

Are you excited about dedicated mobile AI chips? Share your thoughts in the comments.

The post Editorial: The invasion of mobile AI chips is nigh appeared first on AI News.