You are using deep learning on your mobile!
Deep learning is actually a behind-the-scenes helper to make our life more convenient! It’s based on multi-layer feature extractions to know what the data contain. This cloud be applied on various data types, such as figures, voices, and texts. To use a simple deep learning model which contains just two or three hidden layers is unlikely to achieve a judgments just like a human. Therefore, a deep-layered model is used to learn feature maps layer-by-layer, the deeper layer could have a high level expression to make a decision which is expected to be similar as a human does. There are several popular apps that has used deep learning for classification, detection etc. Let's take a look at the hot apps.
- Object detection : Tik Tok、17 Live
- Photo classification : Album、Facebook、Google Album.
- AI Special effects: Meitu-XiuXiu、Instagram
- Object detection
Object detection models are used in several mobile applications, such as Line and Instagram. Once you take a photo, the model will detect your face first and then put special effects on your photo. In the field of the applications, face and scene recognition can use classification methods to enhance photo quality. As Picai APP shows, you can take a high quality photo with filters which are chosen by the APP detection.
From: Picai - Smart AI Camera in GooglePlay
There are three common computer visions problems which could use deep learning techniques to solve and those are classification, detection and segmentation.
First, we will describe what classification is. The classification method can calculate an input image into probabilities distribution of the event over ’n’ different events. This is the simplest and most basic image processing task using deep learning techniques. And, this is also the first breakthrough for achieving a large-scale application. As Google Lens APP shows, you recognize an animal via taking a picture.
From: Google Lens in GooglePlay
Detection is another computer vision method. It focuses on the specific objects it learned. The model calculates the input image into a list of where the objects are and what they are simultaneously. With this technique, you can perform object counting in an image. As object counter by Camera APP shows, you can predict the coin counts via your camera.
Source: Object Counter By Camera in GooglePlay
The third one is segmentation. It includes semantic segmentation and instance segmentation. The former is an extension of the pre-background separation, which requires separation of image parts with different semantics, while the latter is an extension of the detection task, which requires describing the outline of the target (more detailed than the detection frame). Segmentation is a pixel-level description of an image that gives each pixel category (instance) meaning and is useful for understanding more demanding scenes, such as road and non-road segmentation in AI driving.
- Conclusion
Today, almost everyone has a mobile phone. With deep learning applications and mobile device computing capabilities growing, we can experience lots of new and convenient applications easily.
Reference:
● https://chenhh.gitbooks.io/multiperiod_portfolio_optimization/content/ml/deep_learning_intro.html
● https://panx.asia/archives/53209
● https://www.jianshu.com/p/8633d6896255
● Google 相簿人臉辨識搜尋支援台灣