Lab Assignment 6 - GeoSnipes/Big-Data GitHub Wiki

Sub-Team Members

5-2 15 Naga Venkata Satya Pranoop Mutha

5-2 23 Geovanni West

Create the Following Android Applications:

TensorFlow Server REST API & Web Client

Image Classification using Tensorflow Lite

Compare and contrast the results from all the three approaches

Dataset

The dataset used in this assignment is 10 classes of different foods. The classes are images of 10 different categories are as follows:

caesar_salad
caprese_salad
donuts
dumplins
frenchfries
greek_salad
guacamole
hotdogs
risotto
sushi

Each class contains about 300 images which was used to train the model.

Android Client connecting to Tensorflow Server Using REST API

Web Server Results

Response received from Web Server

The server code was unchanged from Lab 5.

New code was created in order to create a connection between the Android app and the Tensorflow Web Server. The app takes a picture of the food and then converts it into base64 (used by the web server) and wrapped into a JSON formatted object. It then forwards this data to the server using the POST method.

The server then takes this data and runs Tensorflow to get a prediction and replies back with its best result.

Tensorflow Lite Application

After getting our model converted we tested it and it successfully worked.

It was noted that the longer the time given for the application to focus on the image the better the prediction probability became. For example, on first look it gave a 62% chance that it was french fries, then around 4 seconds later it had raised to 94%.