FoodSnap.ai Challenge – Implemented with Deep Learning

[devsk18]

Demo

foodsnap.ai-preview.mp4

Solution

Developed a web application that enables users to detect and analyze the nutritional content of food from images. The app provides insights into daily intake statistics and a history of past detections. It leverages two deep learning models:

• A food classification model (MobileNetV2, 87% test accuracy) trained on a Kaggle dataset
• A weight estimation model (CNN-based, RMSE: 131.78, R²: 0.29) trained on data from HuggingFace.

Additionally, implemented a Flask-based API for serving real-time model inferences and built a web scraping script to extract nutrition data and store it in CSV format for database seeding.

Limitations

• Weight estimation has ±131g error; accuracy can improve with a better dataset and model (working on it).
• Food classification is trained on limited items; expanding the dataset will enhance performance (working on it).
• The model can classify but not count multiple instances of the same food.

Technologies Used

Python

• TensorFlow & Keras (Model Training)
• Flask (Inference API)
• Pandas, NumPy (Data handling)
• requests (data fetching)

PHP

• Laravel
• FilamentPHP
• Livewire
• HTML/CSS

MySQL

• User data storage

Google Colab

• Model training

---

Installation and setup instructions are provided in the README.md file