The trend for Mobile AI Application

    Since 2017, Mobile devices are enabled with AI capabilities by adopting advance deep learning model and powerful chips. Huawei Kirin970, Apple Bionic A11, and Mediatek P60 provides a dedicated AI processor to achieve high performance and power efficient for mobile phones. Mobile phone vendors also develops several AI application to utilize this AI processor.


  • Camera Application

    The camera is the most important feature of mobile phones, with AI technology, the camera App became smarter with more detection tolerance on a various angle, light conditions. Thus provides better experience while taking a photo.

    With AI photo classification, the album can classify the picture with its content, person, location, event and the user can organize photo album easily. Face detection provides precise adjustment of focus and exposure while shooting a photo. Scene detection and face landmark detection provides better picture quality for landscape and portrait photos. Semantic segmentation performs pixel level classification, thus the device can adjust the image quality of the specific object in the picture.

From:Google Photos in GooglePlay 

  • Real-time Video and AR application 

    Real-time video and Emoji/AR application require higher computing capability because it needs to finish the processing of an image frame within 33ms(30fps), the latency of AI processing became very important for such real-time application.

    Besides performance/latency, the accuracy is critical for some application such as beautifier for real-time video conference, if the facial landmark is not precisely detected. The cosmetic effect will be placed at the wrong position.

    Next generation AI processor with performance and high accuracy would make real-time video/AR application better experience.


  • The voice assistant 

    Siri was introduced with iPhone 4s in 2011. It performs a simple task in the mobile device and delivers a complex task to the cloud service. Since 2017, iOS provides offline Siri, that means the mobile device is capable of handling the complex voice processing tasks. It’s accelerated by an AI processor of a mobile phone but the experience still has improved to fulfill user expectation. 


  • System Management and Context-Aware assistant

    The mobile phone can manage itself by using AI technology. Analysis of user usage model and predict user needs, then tuning the system task and workload to improve battery life. The device also can provide information, suggestions and assistance based on user’s location/usage and can predict the user needs. Such capability makes the assistant smarter.


  • Networks for various applications

       Since 2017, AI for Image application is widely in use, such as photo classification and scene detection in camera App. 

With the improvement of neural network algorithms and AI processor, real-time video and voice application would come.


Function

Technology

Networks

Computational Intensity

Portrait Effects

(Bokeh …)

Segmentation

DeepLab v3

,Mask R-CNN

*****(Preview) / ****(Process)

Object Recognition

Image Classification

MobileNet,

ResNet,

InceptonV3

**** (Non realtime task)

Object Detection

Object Detection

Mobienet-SSD,YOLO V2

*****(Preview) 

Translation

Sequence to Sequence

Translation

RNN,LSTM

****

Voice Assistant

Voice Recognition

RNN,LSTM

***

Device Optimization

Sequence Detection

/Classification

LSTM

**


  • Next generation of AI processor

Unlike CPU and GPU, they improve the performance by 15~30%, compared to the previous generation. With huge demand of computing power of novel neural networks, AI processors increases the performance by several times. Kirin980 is a quad capability processor and performs 2.2 times faster than the previous generation. iPhone A12 Neural Engine provides 8 times capability than A11. With the greater AI capability of a mobile phone, more advanced AI applications will be enabled, such as picture/video quality improvement, video analysis, and real-time video effects.



Huawei NPU

Apple Neural Engine(Bionic)

2017

Kirin970(NPU)

A11 (Dual Core)

2018

Kirin980(Dual Core NPU)

A12(8 Core)

Growth

4x capability,2.2x faster

8x capability


Note: Information from their product launch event.


Related Posts

Comments

Write Comment