Food
Calorie
Estimation
Team Members
Mukesh Reddy-063
Maneesha-138
Suneel -139
Supervised by
Dr.Sibendu Samanta
Objective
The objective of this project is to develop an
intelligent system that can automatically recognize
food items from images using Convolutional Neural
Networks (CNNs) and accurately estimate their
calorie content. This system aims to assist users in
tracking their dietary intake, promoting healthier
eating habits, and supporting personal health and
fitness goals through a fast, user-friendly, and
reliable application.
Image Recognition
Convolution Layers – CNNs apply filters to small sections of an
image, detecting patterns like edges, textures, and shapes.
Pooling Layers – These simplify the image by reducing its
dimensions while keeping the key features intact.
Activation Functions – Non-linear transformations help the
network understand complex patterns beyond simple lines.
Feature Extraction – Layers progressively refine features,
detecting objects or structures within the image.
Methodology
Image Collection
Collect a dataset of food images with labels and preprocess them to make suitable for model
input
Model Selection
Used CNN to recognize food items
Model Training
Split the data and train the model to learn features and classify food images
Food Recognition Model
Using pre-trained model for image classification
Calorie Estimation
Once the food is recognized, app matches it to its nutritional database and estimate the calories
Why CNN?
CNNs are specialized for visual data—they're designed to
capture spatial hierarchies, patterns, and textures from images.
Food recognition is highly visual (color, texture, shape), and
CNNs excel at extracting meaningful features from these
elements.
They enable advanced tasks – Beyond recognition, CNNs
support powerful extensions like segmentation and volume
estimation, which are essential for accurate calorie prediction.
Advantages
One can easily complete his/her daily calorie intake
Beneficial for people having diabetes, heart problems or
fitness plans
No need to search manually—just upload a food image and
instantly get the food type and nutrition data
Helps users control portion sizes, stay within dietary limits,
and monitor progress over time
It encourages healthy eating habits by helping users make
informed food choices based on nutritional data
Challenges
Calorie Estimation Complexity
Caloric content varies by recipe, cooking method, and ingredients.
Food Variability
Same dish can look very different (e.g., homemade vs. restaurant).
Mixed dishes and overlapping foods make it hard to segment individual
items.
Model Generalization
Model might perform well on known data but struggle with unseen food
items. Domain adaptation required for different regions/cultures.
Key Functions
classify_image(image_path)
Classifies the image and predicts the food label using a pre-trained
model via Hugging Face
Confidence_Threshold
Confidence Threshold is a minimum confidence score that the
model must have in its prediction to be considered reliable
Get_calories(food_name)
Retrieves nutritional information based on the food label predicted
by the model.
Includes error handling for API failures
Error_handling
In our code we used try and except method to clear the errors
Accuracy Graph
The following graph is a comparison of the accuracy between the values provided by Google and generated by our website.
Conclusion
Our project integrates advanced AI to deliver an efficient and user-
friendly solution for food identification and nutritional analysis. By
using deep learning and external APIs, it aligns with the growing
demand for health-conscious technologies. Based on our idea, a
future device can be developed to detect food and estimate
nutritional values based on volume and recognition
Thank You
