🍽️ FoodCamp

Food waste reduction platform — restaurants post surplus food with AI-detected item names, nearby food camps browse a live map, claim portions, and mark pickups.

FastAPI · React · PostgreSQL · ResNet-50 · Leaflet / OpenStreetMap · AWS S3

---

How it works

Restaurant uploads food photo
        │
        ▼
ResNet-50 (Food-101, fine-tuned)
→ auto-detects food name + suggests quantity
        │
        ▼
Listing published on live map (Leaflet/OSM)
        │
        ▼
Camps browse map → claim a portion → set pickup ETA
        │
        ▼
Auto-release: unclaimed portions return to pool after 4 hours
        │
        ▼
Camp marks pickup complete → listing closes

---

Features

Role	Capabilities
Restaurant	Upload food photo, confirm AI-detected name & quantity, manage Active / Past listings
Camp	Browse available food on map or list view, claim portions with pickup ETA, mark as picked up

• AI food scan — ResNet-50 fine-tuned on Food-101 (101 food categories). Set MOCK_SCAN=true to skip the model in dev.
• Partial claiming — multiple camps can claim portions of the same listing until fully claimed
• Auto-release — claims auto-expire after 4 hours, returning quantity to the pool
• Geocoding — restaurant location resolved via Nominatim (OpenStreetMap) — no API key needed
• Image storage — AWS S3 in production, falls back to local /tmp in dev

---

Tech Stack

Layer	Technology
Backend	Python 3.9+, FastAPI, SQLAlchemy, Pydantic v2, Uvicorn
Auth	JWT (python-jose), BCrypt (passlib)
Frontend	React, Leaflet / react-leaflet, OpenStreetMap
Database	PostgreSQL (psycopg2)
ML	PyTorch, torchvision, ResNet-50 (Food-101 fine-tune)
Storage	AWS S3 (boto3) — local /tmp fallback for dev
Geocoding	Nominatim (OpenStreetMap) — no API key

---

Project Structure

foodcamp/
├── app/
│   ├── main.py                 # FastAPI app, CORS, auto-release background task
│   ├── database.py             # SQLAlchemy engine + session
│   ├── config.py               # Pydantic settings from .env
│   ├── models/                 # SQLAlchemy ORM models
│   │   ├── user.py             # Base user
│   │   ├── restaurant.py       # Restaurant profile
│   │   ├── camp.py             # Camp profile
│   │   ├── food_listing.py     # Surplus food listing
│   │   └── claim.py            # Camp claim on a listing
│   ├── routes/
│   │   ├── auth_route.py       # Register / login → JWT
│   │   ├── restaurant.py       # Upload listing, manage listings
│   │   └── camp.py             # Browse map, claim, mark pickup
│   ├── services/
│   │   ├── food_scan.py        # ResNet-50 inference (or mock)
│   │   ├── geocode.py          # Nominatim geocoding
│   │   ├── s3_service.py       # AWS S3 upload / local fallback
│   │   └── auth_service.py     # JWT issue + verify
│   ├── repositories/           # DB query layer
│   └── ml/
│       ├── loadModel.py        # Load fine-tuned ResNet-50 weights
│       └── preprocess.py       # Image transforms for inference
├── scripts/
│   └── train_food101.py        # Fine-tune ResNet-50 on Food-101 (~1–2 hr GPU)
├── frontend/                   # React app (map + listing UI)
├── requirements.txt
└── .env.example

---

Quick Start

Prerequisites: Python 3.9+, PostgreSQL, Node 18+

# 1. Clone & install
git clone https://github.com/spandey1702/FoodCamp.git
cd FoodCamp
python -m venv venv && source venv/bin/activate
pip install -r requirements.txt

# 2. Configure
cp .env.example .env
# Edit .env — set DATABASE_URL and SECRET_KEY at minimum

# 3. Run backend  (Swagger docs → http://127.0.0.1:8000/docs)
uvicorn app.main:app --reload

# 4. Run frontend
cd frontend && npm install && npm start
# → http://localhost:3000

Environment Variables

DATABASE_URL=postgresql://user:password@localhost:5432/foodcamp
SECRET_KEY=your-secret-key

# Set true to skip ML model during development
MOCK_SCAN=true

# AWS S3 (optional — local /tmp used if omitted)
AWS_ACCESS_KEY_ID=
AWS_SECRET_ACCESS_KEY=
AWS_REGION=us-east-1
S3_BUCKET_NAME=foodcamp-images

# Auto-release window (seconds between checks, default 30 min)
AUTO_RELEASE_CHECK_SECONDS=1800

ML Model (optional — skip if MOCK_SCAN=true)

# Downloads Food-101 dataset and fine-tunes ResNet-50
# Requires GPU — ~1–2 hours. Saves weights to app/ml/food101_resnet50.pth
python scripts/train_food101.py

---

API Reference

Method	Endpoint	Description
POST	/auth/register	Register as restaurant or camp
POST	/auth/login	Login → JWT
POST	/restaurant/listings	Create listing (upload photo → AI scan)
GET	/restaurant/listings	My listings (Active / Past)
DELETE	/restaurant/listings/{id}	Remove listing
GET	/camp/listings	Browse all available listings (map data)
POST	/camp/listings/{id}/claim	Claim a portion with pickup ETA
PATCH	/camp/claims/{id}/pickup	Mark claim as picked up
GET	/camp/claims	My claims

Notes

• Claims auto-release after 4 hours if not picked up — quantity returns to the listing pool
• Multiple camps can claim portions of the same listing simultaneously until fully claimed
• Restaurant location is geocoded via Nominatim (OpenStreetMap) — no paid API key required
• The ML model supports 101 food categories from the Food-101 dataset; set MOCK_SCAN=true for instant dev feedback

---