Week 1 — DeepNutrition: Be Aware Of Your Macronutrients
Facilitating the dietary assesment process by detecting multiple foods from a single image
The Problem
Hi,
Last year when I was a sophomore, I became interested in bodybuilding and I applied to gym. As I am new in this area, my friend Ali told me that I should definitely do my dietary assesment. He told me that how he managed to get in form by using an android app which helps him to follow his diet.
He motivated me so much that I immediatley downloaded the app and started recording everything I eaten. It went well. It motivated me to achieve my calorie goal each day. And it made me more conscious because sometimes we can underestimate the calories of foods. I’ve seen so many people get into shape just by doing diet reviews.
Day after day I started to skip some meals because of laziness. It went harder and harder to type everything I have eat, search them find them one by one. And eventually I defeated by lazyness and ended to use that app.
I have seen that people who have enough willpower to continue using this application have reached their goals. And I thought that, What if it was a easier way to use that app ? I shared this question with my friend Muhammed and after a short brainstorm we came up with an idea. We can use the power of Computer Vision!
All the user had to do was photograph the table before dinner. Then we would process that image and identify all the food he\she ate from that photo. It has to be just that simple. By that way the effort to use this app will decrease roughly from (~72 second/meal) to (~9 second/meal). But that was just the ultimate goal and just an idea. We wasn’t know how to build such a thing. We wasnt even know if it is possible to do. And we decided to wait untill we took our machine learning course in college.
The Project
This year we were expected to do a semester project in our BBM406 machine learning course at Hacettepe University. We Ali Kayadibi, İsmet Okatar and Muhammed Aydoğan are team in this project. We thought a lot and came up with different ideas to do.
That thinking process fascinated us by showing the amount of things can be automated by computer vision
Then we remembered our last year’s problem and decided to do it.
Model
Our ultimate goal is to detect multiple foods from a single image. We will choose algorithms like SSD and YOLO and for model maybe Inception, ResNet or VGGNet . After we finish our research we will run tests and choose the best one.
Dataset
There is plenty of data for foods. One of them is https://www.vision.ee.ethz.ch/datasets_extra/food-101/
Our ultimate goal is to recognize all possible foods. However, since it is a term project, we will start small, maybe 10–20 classess and we will increase the number of classes if we can.
Fact
In 2015 107 million children and 603 million adult were obese. Having a high body mass index accounted for 4 million deaths in 2015. And more than two thirds of these were from heart disease.
Conclusion
We strongly believe that dietary assesment plays an important role in the fight against obesity and related diseases. Therefore, we think that facilitating the assesment process can make a great contribution! Who knows maybe the increased data flow about daily nutrition intake can lead to many applications that will make our lives better.
