Pyrid

Inspiration

Among many of our friends, some are desperate to lose their love handles while others yearn to put on some weight. This desire led most of us to research healthier eating habits to balance different aspects of dietary needs. But what if you didn’t need to spend hours studying what foods to consume, and could instead simply give an AI all the information it needs to help you elevate your weight transformation journey to another level?

What it does

Our project helps everyone find a quick and convenient way to eat healthier, even those who lack a comprehensive understanding of dietary needs! It’s a website that collects what you’ve eaten and creates an in-depth visualization of it, allowing users to track their daily intakes and learn how they could further improve their eating habits.

How we built it

Being food connoisseurs ourselves, we were able to utilize much of our knowledge of nutrition and dietary habits to gain a comprehensive understanding of what we wanted our website to do. We used an object detection system to capture the foods and ingredients that users have consumed and paired the collected data with a large database of nutritional values. From there, we used a custom grading system to determine what nutrients users would need to prioritize and give further analysis into how they could improve in the future.

Challenges we ran into

Creating this tool required extracting a picture into its simplest subcomponents with an accurate estimated weight. Some solutions required downloading huge datasets, while others returned the subcomponent weights in five different measurements. There were many options we tried, such as OpenCV and Imagga, that proved ineffective for our project. In the API we finally chose, many food queries were categorized into completely different items, like corn oil instead of corn on the cob. To solve this, we matched keywords and determined a closeness score to find the best match.

Accomplishments that we're proud of

Translating a basic image into the essential molecules that fuel your body was a difficult yet rewarding task. Researching numerous scientific articles to develop an algorithm to obtain the most accurate estimates for daily macro/micronutrient intake from basic user data (weight, height, and sex) was also a major accomplishment for us.

What we learned

We learned how to turn a rushed image into a precise ingredients list, specific recommendations based on pure nutrient needs, and above all else, the importance of maintaining a healthy diet. Besides becoming experts in micronutrients, we also learned to integrate an immense government dataset to solve real-world scenarios applicable to everyone.

What's next for Pyrid

The largest estimation in Pyrid is the weight of the food. From lighting to an off-angle, countless factors can alter the perceived weight of a food item. Instead of requiring any manual input, we plan to promote uploading multiple angles of the same food from differing angles, aiding in the weight estimation and thus the nutritional breakdown. Furthermore, the API used to obtain the nutrition information can be further enhanced, adding a wider variety of food including more complex or niche items.