CSE535 Mobile Computing - Offloading Android Application
Title - Android Image Classification and Offloading
Background and Motivation - Computational offloading is a critical technology for emerging edge computing and Internet of Things (IoT). Offloading is a technique to increase mobile device’s capability by dividing the problem and migrating the computation to other more resourceful devices or servers located nearby. This improves the application performance as well as serves as a solution for the limited resources of IoT devices. This project demonstrates computational offloading in edge computing by offloading the task to multiple flask servers, thereby increasing speed and accuracy, as well as reducing the computational cost.
Goal of the project - The goal of this project is to use an Android application to partition handwritten digit images into 4 parts and send them to 4 different servers for classification. The classification is done by training a machine learning model (CNN) on the MNIST dataset in order to predict the image parts received. Following the classification of the image parts on the 4 servers, the results along with the accuracy of classification is sent back to the master mobile. The master mobile then selects the classification based on highest accuracy, and places the digit image in the relevant folder on the mobile internal storage.
Project files -
- The main.py is used to create and run flask servers on different devices. Each of these servers use an already trained model 'mymodel.h5' in order to carry out the task of classification of handwritten digits.
- The Train_model_and_Save.ipynb file is used to save the model in .h5 format.
- As part of the initial stages of project development, a single Flask server was built to carry out the task of classification as well as prediction of the new input data received on the server. This is compiled as a separate piece of code in 'MobileComputing-Assignment2 - Single Flask server classification' and can be run independently as a server.
Install -
- Start the flask server on 4 different laptops by running the main.py file, and run the ngrok tool to expose the localhost to the public domain.
- Install the application on the master android mobile.