Most diet trackers are a dime a dozen, but what if you could track your eating habits and find low-cost recipes tailored to your needs--simply by taking a picture? NutriTrack makes diet tracking more convenient, memorable, and manageable by vastly simplifying user input, as well as implementing cost comparison for the cook on a budget. NutriTrack is inspired by a current lack of diet trackers that are socioeconomically inclusive and convenient.
What it does
NutriTrack allows users to log their nutrition by simply uploading pictures of food to their dashboard. After using classification to determine the user's meals, NutriTrack analyzes diet information over time to recommend general diet alterations and healthy recipes. In addition to user data, recipes are organized and reported based on low-cost alternatives to more expensive health foods. NutriTrack accesses pricing data and organizes recipes regarding ascending price of ingredients and descending nutritive value, making NutriTrack an accessible resource for low-income families pursuing healthy eating.
How we built it
NutriTrack is a web app developed using IPython and Django. Image classification is performed using a deep convolutional neural network trained on the Food-101 dataset. We utilized a 16 K80 GPU instance over Amazon AWS to achieve state of the art results on this dataset in under 8 hours. This network helps us to conveniently serve input to customer profiles regarding their nutrient intake. Upon doing this, we used the nutritionix API to access recipes and nutrition information for imported items. We then created an algorithm to sort the recipes based on individual requirements for ten different dietary values (calories, fat, vitamin A, vitamin C, calcium, carbs, protein, sodium, iron, and sugar). Recipe sorting also includes a pricing filter to suggest cost effective recipes; this is done using Walmart grocery prices from compare groceryprices.org.
What's next for NutriTrack
We would like to expand the scale of NutriTrack's datasets in order to generate more accurate and pertinent information for users. This would include the creation of more localized and comparative pricing sources, more recipes, the inclusion of micronutrients, and more classes in the classification algorithm.