MosesOnerhime/calvision

A web-based African food recognition and calorie estimation system. Upload a meal photo and get instant nutritional breakdown. Built with React + TypeScript (frontend), Django REST Framework (backend), and a custom-trained TFLite model for Nigerian/African cuisine recognition. # CalVision - Food Recognition & Calorie Estimation CalVision is a full-stack app for logging meals from photos. Users can upload a meal image, receive Nigerian-food segmentation/classification with estimated calories/macros, and save the result to their meal history. ## Quick Start (Docker) ```bash # 1. Clone the project git clone && cd calvision # 2. Copy env file and add your USDA key if you have one cp backend/.env.example backend/.env # 3. Start everything docker-compose up --build # Frontend -> localhost # Backend -> localhost # Admin -> localhost ``` ## Manual Setup ### Backend ```bash cd backend python -m venv venv && source venv/bin/activate # Windows: venv\Scripts\activate pip install -r requirements.txt cp .env.example .env # Edit with your values. python manage.py migrate python manage.py createsuperuser python manage.py runserver ``` ### Frontend ```bash cd frontend npm install echo "VITE_API_URL=localhost" > .env npm run dev -- --port 3000 ``` ## USDA API Key Get a free key at: api.nal.usda.gov Add it to `backend/.env` as `USDA_API_KEY=your_key` ## Environment Variables | Variable | Description | |----------|-------------| | `SECRET_KEY` | Django secret key | | `DEBUG` | True for development | | `DATABASE_URL` | Database URL. Defaults to local SQLite if omitted. | | `USDA_API_KEY` | USDA FoodData Central API key | | `ALLOWED_HOSTS` | Comma-separated Django host allowlist | | `CORS_ALLOWED_ORIGINS` | Comma-separated frontend origins allowed by the API | Frontend variables: | Variable | Description | |----------|-------------| | `VITE_API_URL` | Backend API origin, for example `localhost` | ## AI Pipeline CalVision uses a YOLO instance segmentation model first. It detects visible food regions, returns bounding boxes and segmentation masks, and the backend draws the colored AI output overlay shown on the results page. The tr …