- Trending Categories
- Data Structure
- Operating System
- MS Excel
- C Programming
- Social Studies
- Fashion Studies
- Legal Studies
- Selected Reading
- UPSC IAS Exams Notes
- Developer's Best Practices
- Questions and Answers
- Effective Resume Writing
- HR Interview Questions
- Computer Glossary
- Who is Who
How to apply machine learning in an android app?
Nowadays there are more intelligent recommendation systems, AI embedded cameras on smartphones for auto-focus and auto-face detection, advanced chatbots, and ai voice assistants like Siri, Alexa, and Google Assistant. People have started creating apps like ChatGPT, which can handle advanced problems like interviews. However, fascinating this may seem but to deploy an app with such capabilities one needs to have in-depth knowledge of machine learning and deep learning algorithms, with access to a huge variety of resources and datasets, which sometimes is impossible to attain by single individuals. So a proper group of people is needed to access and manage such huge amounts of resources and to deploy machine learning into apps with huge utilities. Still, it is quite possible to implement some basic and pre-trained or pre-implemented machine learning algorithms into our apps.
ML Kit − This kit works on both Android and iOS, platforms. These APIs are already optimized for mobile and don’t require much expertise in Machine Learning. The key advantage of ML Kit is that developers can run machine learning models on the device itself, instead of sending data to a remote server for processing. This helps in improving the privacy of the user data, as the data need not leave the device, moreover, this also helps the developers to provide faster data processing and produce real-time results. Some of the APIs include −
Vision APIs −
Barcode Scanning − Barcodes have become convenient for passing information through devices. Using this API one can easily scan and fetch the information embedded in a barcode, and the app can respond to the data according to the type of data received. The barcode must be in linear formats (Codebar, Code 39, EAN-8, EAN-13, ITF, Code 93, Code 128, etc.) or 2D formats (QR Code, PDF417, Data Matrix, Aztec).
Face Detection − Now, most certainly you have used Snapchat’s face beautification features, Instagram's ‘What Pokemon are you?’, etc. This type of feature can be implemented using face detection. Note: this only detects the faces, and cannot identify whose face it is (face recognition is not done). It can also be used for image manipulation on a particular person in any recorded video.
Face Mesh Detection − For this, you need to have a selfie-like picture, over which the ML model will draw a bounding box after detecting the face. It generates AR effects on faces in streaming videos. This can be used to get the 468 3D points and triangle info.
Text Recognition − Hopefully, all have used and have some idea of text recognition, so this is used when you click a picture of some article paste it into Google, and ask him to convert it into your mother language. But there are limitations, this works only upon the Latin-based languages, as the ML kit supports only Latin Script. Also, it detects the words, paragraphs, lines, and each and every element of the context. Also, it can identify which language you are typing or feeding.
Image Labelling − We can detect and extract information about the various objects or entities present in an image across a broad group of categories. We can also customize these features using the TensorFlow Hub, or our own models trained with TensorFlow, AutoML Vision Edge, or TensorFlow Lite Model Maker. ML Kit extracts the labels from the TensorFlow Lite model and Provides them as a text description.
Digital Ink Recognition − with the help of this API we can recognize handwritten text through the screen of our device. It can also be used to classify gestures on a digital surface like your phone’s screen, in hundreds of languages, also it can classify sketches. With the help of this, we can write our language with ease on the screen of the phone rather than cumbersome typing through the virtual keyboard.
Pose Detection − This API allows the developers to add features to the app which work with the poses of the body.This can be used to identify any person’s position in real-time from a video or an image. This API could be pretty useful in health and fitness apps to detect your motion.
Selfie Segmentation − Ever used background remover to place yourself in images that were never actually taken, or to show yourself in locations where you have never been. This API allows you to do that. With the help of this API, developers can easily separate the background from the people or chosen objects in the image. It also works on any video in real-time.
Natural Language APIs −
Language Identification − This API helps the developers to identify the language of the input text or string, as users generally don’t give their language information while doing a Google search, or typing any text.
Translation − This API dynamically translates any given language into a different output language as desired by the user. Here dynamically means that the app can dynamically download and manage language packs, hence keeping the device storage requirements low. Smart Reply: This API allows apps to suggest autonomous replies to any text messages received on your device. It generates reply suggestions based on the full context of a conversation.
Entity Extraction − Modern smartphones suggest texts over a small bar above the virtual on-screen keyboard, ever wondered how this happens? It’s the functioning of this API. While typing any text this API identifies the words or entities used most often, extracts them, and suggests them as shortcuts based on the context you are typing. Some of the entities include address, date-time, email addresses, flight number, money/currency, phone number, URL, etc.
However, cool this may sound till now still there are certain limitations to these pre-trained machine learning models they may not be accurate and lack performance in some use cases of your apps.
Android’s Custom ML Stack − If you want heavy control and to deploy your own machine learning models, then you can use this. This custom ML Stack is built upon TensorFlow Lite and Google Play Services, so it can provide the most needed essentials to deploy high-performance ML features. From the diagram above it is clear how TensorFlow Lite can be used to control the performance of your apps at various levels.
For ML Runtime − You can use TensorFlow Lite via Google Play services to run high-performance ML inference in your application.
TensorFlow Lite Delegates − If you are struggling with your app's performance then you can use TensorFlow Lite Delegates which is distributed via Google Play Services to accelerate your ML model. Accelerating your ML model means providing faster solutions with pinpoint accuracy. This acceleration takes place over specialized hardware like GPU, NPU, or DSP. This reduces the latency and provides a fluid user experience to your users by accessing the advanced computing capabilities of the user’s device.
But an important catch here is that you need to change your, initial model, into the TensorFlow Lite model (.tflite). And your new model should be very very different from the pre-implemented TensorFlow Model. The updates of such apps are managed by Google Play Service and the updates are provided to all the users as soon as any new changes are made. If you want to avoid this and give specific updates to specific users, you can use Firebase Remote Config and Firebase A/B testing.
Kickstart Your Career
Get certified by completing the courseGet Started