Automated Nutrition Tracking
Simplifies calorie counting by removing the need for manual input.
Technologies used
Python
Caffe
OpenCV
Motivation
The objective was to develop a more efficient and accurate solution for calorie counting by using deep learning and computer vision. Unlike current calorie tracking applications that rely on manual data entry, Azati aimed to automate the process of estimating calorie content through the analysis of food images.
Main challenges
Challenge 01
Training Data Quality
One of the key challenges was ensuring a sufficient amount of high-quality training data. Different products and cooking methods resulted in varied calorie counts, and Azati had to collect diverse images of dishes to account for all possible variations. For example, the calorie content of a boiled egg differs from a fried one, and this needed to be captured in the data.
#1
Challenge 02
Image Quality and Angle Issues
The AI algorithm struggled with inconsistencies in lighting, angles, and dish sizes. To overcome this, Azati had to account for situations like partial food images or varying plate sizes. Manual adjustments were made to compare and calculate data accurately, ensuring that the algorithm could recognize and estimate calorie content correctly.
#2
Our approach
Market Research
Azati began by analyzing existing software in the market to understand basic operational principles and identify areas for improvement. This step ensured the team could build on existing technologies and enhance the user experience.
Algorithm Development
After studying the necessary documentation and tools, Azati’s team developed a custom algorithm to process food images and estimate calorie content. The algorithm was designed to handle complex dishes and multiple ingredients.
Data Collection and Preparation
Azati collected and prepared high-quality images of dishes for training the AI model. The process involved gathering images from open sources and creating new ones manually to ensure a wide variety of food types and cooking methods were represented.
Machine Learning Training
Once the images were prepared, Azati trained the machine learning model to automatically recognize and categorize food components. The team manually mapped categories, ensuring the AI could distinguish between different parts of a dish, such as garnish and meat.
Template Creation
For each group of food components, Azati created templates that helped the system process the images with maximum accuracy. These templates were used to analyze all images within their respective categories, ensuring consistent results.
Iterative Refinement
The system underwent multiple rounds of processing to refine the data. Each round improved the accuracy of the calorie estimations, resulting in a highly functional prototype.
Facing the same challenge?
Bring your complexity. We'll bring the plan. Select a convenient slot to start a conversation with our experts.
Schedule a callSolution
01
AI-Based Food Recognition Module
The prototype uses AI and computer vision to automatically recognize food components in images, including complex dishes with multiple ingredients. The system estimates the calorie content of each component and calculates the total calories for the dish.
Key capabilities:
- AI-driven recognition of individual food components
- Supports complex dishes with multiple ingredients
- Accurate calorie estimation per item
02
Visual Calorie Breakdown Module
The application visually displays the calorie count for each individual component of a dish, using frames to circle the recognized food items. This provides users with an easy-to-understand breakdown of their meal's nutritional content.
Key capabilities:
- Visual highlighting of individual ingredients
- Calorie breakdown per food component
- Improves user comprehension and engagement
03
Manual Mapping and Template Module
To maximize accuracy, Azati’s team manually mapped various food components, such as separating the garnish from the meat. This ensures that the system delivers precise calorie counts even for dishes with multiple ingredients.
Key capabilities:
- Manual labeling and categorization of food items
- Template-based recognition for consistent accuracy
- Reduces errors in multi-ingredient dishes
04
Real-Time Processing and Reporting Module
The AI-powered system processes images in real-time, providing users with an instant estimate of their meal’s calorie content, making it faster and more efficient than traditional manual input methods.
Key capabilities:
- Instant calorie estimation from food images
- Real-time feedback for users
- Scalable processing for thousands of dishes
Results & business impact
Faster and More Accurate Insights
Processing time decreased by 60–80%, and calorie estimation accuracy improved by 85–95%.
Improved Model Performance
AI accuracy and recognition improved by 40–55%, ensuring reliable results for complex dishes.
Applicable Across Industries
Healthcare, fitness, catering, agriculture, and daily nutrition benefit from automated food recognition.
Enhanced User Experience
Visual breakdown of meals makes nutrition tracking intuitive and actionable.
Scalable and Extensible
The system can handle thousands of dishes and ingredients, ready for future enhancements or integration with health platforms.
Data-Driven Decision Making
Enables users and organizations to monitor diets, improve meal planning, and support health-related insights.
Cost and Time Efficiency
Reduces time spent on manual tracking while increasing accuracy and consistency.
