Lab Assignment 5 - GeoSnipes/Big-Data GitHub Wiki
5-2 15 Naga Venkata Satya Pranoop Mutha
5-2 23 Geovanni West
- Write a TensorFlow program for the following Task:
- 1. UseCNN for your project dataset
- 2. Use Inception Model on your dataset
- 3. Use FLASK for creating Web Application for your dataset.
The data set used in this assignment is 5 classes of different foods. The classes are images of 5 different categories as follows: applepie, babybackribs, beefcarpaccio, caesarsalad, and chickenwings. Each class contains about 1000 images.
![](https://github.com/GeoSnipes/Big-Data/raw/master/lab_assignments/Lab%205/Documentation/CNN%20Accuracy.png?raw=true)
This model may not be appropriate to use with our datatset. It gave a testing accuracy of only 41%. I also tried setting the number of steps/iterations to 2000, it resulted with a 92% training accuracy but the testing accuracy barely increased with a value of 44%.
![](https://github.com/GeoSnipes/Big-Data/raw/master/lab_assignments/Lab%205/Documentation/CNN%20Tensorflow%20cross%20entropy.png?raw=true)
![](https://github.com/GeoSnipes/Big-Data/raw/master/lab_assignments/Lab%205/Documentation/Inception%20Accuracy.png?raw=true)
The inception model provided a much better result of 87.5% testing accuracy
![](https://github.com/GeoSnipes/Big-Data/raw/master/lab_assignments/Lab%205/Documentation/Inception%20tensorflow%20cross%20entropy%20and%20accuracy.png?raw=true)
The model does a good job in predicting the food.
![](https://github.com/GeoSnipes/Big-Data/raw/master/lab_assignments/Lab%205/Documentation/webapp1.png?raw=true)
![](https://github.com/GeoSnipes/Big-Data/raw/master/lab_assignments/Lab%205/Documentation/webapp1flask.png?raw=true)
Gave a 40% probabilty match that the food was caser salad
![](https://github.com/GeoSnipes/Big-Data/raw/master/lab_assignments/Lab%205/Documentation/webapp2.png?raw=true)
![](https://github.com/GeoSnipes/Big-Data/raw/master/lab_assignments/Lab%205/Documentation/webapp2flask.png?raw=true)
Gave a 52% probabilty match that the food was barbecue ribs. Can also motice that chicken wings was a not too far second with 29% probablity match.