Know-Before-You-Eat

An Interactive web application for identifying food names based on the images, providing nutritional facts (For eg: calculating calories of the food you're eating) for diet advice and predicting the recipes based on the predicted food names.

Final Deployed Application Link

Presentation Link

https://drive.google.com/file/d/1RufkM7jekyF-5DaBEeNuWt1R1k_Tdox1/view?usp=sharing

Source Data

1. Food101 Dataset https://www.vision.ee.ethz.ch/datasets_extra/food-101/

2. Nutritional Facts https://www.fatsecret.com/calories-nutrition/ http://ahealthylifeforme.com

3. Recipe https://www.kaggle.com/kaggle/recipie-ingredients-dataset https://en.wikipedia.org/wiki/

Tools/Models Reference

1. Classification/Training Models

   Transfer Learning With MobileNet

   Transfer Learning With VGG16

   KNN & Random Forest

2. Keras Image Data Generator for Image Augmentation

3. Front End Application - HTML, CSS, Bootstrap and Javascript

4. Retrieving Data From Back End : Python (SQLAlchemy and Flask)

5. Missing Link AI - Platform to Run deep learning experiments on hundreds of machines, on and off the cloud, manage huge data sets and gain unprecedented visibility into your experiments. https://missinglink.ai/

Presentation: https://docs.google.com/presentation/d/1UvU7sMPkn2Y5I88bmdoH0Sog3SpyQeSPeVdtSZoHrDk/edit#slide=id.g6ea730b6b1_7_5

Results

1. After fine-tuning a pre-trained MobileNet model achieved about 99.03% Top-1 Accuracy on the Training set and about 73% accuracy on Valid & test data.
2. After fine-tuning a pre-trained VGG16 model achieved about 98.03% Top-1 Accuracy on the Training set and about 70% accuracy on Valid & test data.
3. Using KNN Algorithm achieved at score:0.404 at K=3
4. Using Random Forest Model achieved at score:0.2

Key TakeAways

1. Through application of Various Machine Learning Algorithms - K-Nearest Neighbors, Random Forest Classification and Deep Learning(CNN) Algorithms for image classification we concluded that CNN is the best model for classification of images in our data set.
2. In CNN pretrained models Mobilenet model is the best in terms of both speed and accuracy in our dataset.
3. MobileNet is the best method and quickest way to implement transfer learning for CNN’s.