EcoVision — Waste Classification & Mapping

A full-stack web application that classifies waste from images using deep learning (SegFormer + ResNet18), generates recycling guidance via a RAG pipeline (FAISS + Groq LLM), and plots reports on an interactive map.

---

Features

• Image upload with drag-and-drop
• Waste classification (plastic, metal, glass, paper, other)
• AI-generated segmentation overlay
• RAG recycling guide from a PDF knowledge base (FAISS + Groq)
• Interactive map with historical waste reports (MongoDB Atlas)
• AI chat assistant for follow-up questions
• Manual location entry with geocoding fallback

---

Prerequisites

---

Local Setup

1. Clone / unzip the project

cd ecovision

2. Create a virtual environment

python -m venv venv
source venv/bin/activate      # Linux / Mac
venv\Scripts\activate         # Windows

3. Install dependencies

pip install -r requirements.txt

Note: The segment-anything package installs from GitHub. Requires git to be on your PATH.

4. Configure environment variables

cp .env.example .env

Edit .env and fill in:

• MONGO_URI — your MongoDB Atlas connection string
• MONGO_DB_NAME — database name (default: waste_classification)
• GROQ_API_KEY — your Groq API key

5. Add model weights

Place the following files in the project root:

Download SAM weights from: https://github.com/facebookresearch/segment-anything#model-checkpoints

6. Add RAG knowledge base PDFs (optional)

Place recycling-related PDF files in:

backend/rag/data/

The app will automatically index them on first startup and cache the FAISS index to backend/rag/cache/.

7. Run the server

python -m uvicorn backend.main:app --host 0.0.0.0 --port 5000 --reload

Open your browser at: http://localhost:5000

---

Project Structure

ecovision/
├── backend/
│   ├── main.py                  # FastAPI app & routes
│   ├── db/
│   │   └── database.py          # MongoDB connection
│   ├── models/
│   │   └── inference.py         # SegFormer + ResNet18 + SAM pipeline
│   ├── services/
│   │   ├── processing.py        # Recyclability & cleanliness scoring
│   │   └── rag_service.py       # FAISS retrieval + Groq generation
│   └── rag/
│       ├── data/                # PDF knowledge base files
│       └── cache/               # FAISS index cache (auto-generated)
├── frontend/
│   ├── index.html
│   ├── css/
│   │   ├── style.css
│   │   └── components.css
│   └── js/
│       ├── api.js               # API calls
│       ├── app.js               # Main app logic
│       ├── map.js               # Leaflet map
│       └── ui.js                # UI helpers
├── uploads/                     # Saved images (auto-created)
├── requirements.txt
├── .env.example
└── README.md

---

Environment Variables Reference

---

Tech Stack

• Backend: FastAPI, Python 3.11
• ML Models: SegFormer (HuggingFace Transformers), ResNet18 (PyTorch), SAM (Meta)
• RAG: FAISS vector store, sentence-transformers (all-MiniLM-L6-v2), Groq (llama-3.1-8b-instant)
• Database: MongoDB Atlas
• Frontend: Vanilla HTML/CSS/JS, Leaflet.js
• Geocoding: OpenStreetMap Nominatim (no API key required)