Leaf-Expert

Leaf-Expert is an AI-powered plant health tool that classifies leaf diseases and explains why using Grad-CAM++ and LIME — bridging the gap between deep learning accuracy and agricultural interpretability.

---

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                         Leaf-Expert                             │
│                                                                 │
│  ┌─────────────────┐     REST API      ┌───────────────────┐  │
│  │  React + Vite   │ ◄────────────────► │   FastAPI Backend  │  │
│  │  (Frontend)     │                   │   (Python 3.11)   │  │
│  │  :3000          │                   │   :8000           │  │
│  └─────────────────┘                   └────────┬──────────┘  │
│                                                  │              │
│                              ┌───────────────────┼───────────┐ │
│                              │                   │           │ │
│                      ┌───────▼───┐   ┌───────────▼──┐  ┌────▼──┐ │
│                      │  Trainer  │   │  Predictor   │  │ XAI  │ │
│                      │ PyTorch   │   │  (cached)    │  │GradCAM│ │
│                      │ EfficientNet│ │              │  │+ LIME │ │
│                      │ ResNet|ViT│   │              │  │       │ │
│                      └───────────┘   └──────────────┘  └───────┘ │
└─────────────────────────────────────────────────────────────────┘

Features

Feature	Details
Model Architectures	EfficientNetV2-S/M, ResNet50/101, DenseNet121, MobileNetV3, VGG16/19, ViT-B/16
Training	Two-phase fine-tuning, mixed precision (AMP), AdamW + CosineAnnealingLR, early stopping
Inference	In-process model cache, dynamic input size from config, per-class probabilities
XAI	Grad-CAM++ (pytorch-grad-cam) + LIME — both returned as base64 PNGs
Frontend	React 18 + Vite + TypeScript, dark botanical theme, drag-and-drop upload
API	FastAPI 0.115+, Pydantic v2, async endpoints, full Swagger UI at /docs
Deployment	Docker Compose (backend + frontend + nginx)

Quick Start

Prerequisites

• Python 3.11+
• Node.js 20+
• (Optional) CUDA-capable GPU for faster training

1. Clone & Setup

git clone https://github.com/Purushothaman-natarajan/Leaf-Expert.git
cd Leaf-Expert

2. Backend

cd backend
cp .env.example .env          # edit if needed
pip install -r requirements.txt

# For GPU:
# pip install -r requirements-gpu.txt

uvicorn app.main:app --reload --port 8000

API docs: http://localhost:8000/docs

3. Frontend

cd frontend
npm install
npm run dev

UI: http://localhost:3000

4. Docker (both services)

docker-compose up --build

---

API Reference

Endpoint	Method	Description
/health	GET	Server health + PyTorch / CUDA info
/data/prepare	POST	Split raw dataset into train/val/test
/train/start	POST	Launch background training job
/train/status/{job_id}	GET	Poll live training metrics
/predict/	POST	Classify a leaf image
/predict/explain	POST	Classify + Grad-CAM++ + LIME explanations

Full docs: docs/api_reference.md

---

Repo Structure

Leaf-Expert/
├── backend/              # FastAPI ML service
│   ├── app/
│   │   ├── api/          # Route handlers
│   │   ├── core/         # Config + logging
│   │   ├── schemas/      # Pydantic models
│   │   ├── services/     # Business logic (data, trainer, predictor, explainer)
│   │   └── main.py
│   ├── tests/
│   └── requirements.txt
├── frontend/             # React + Vite + TypeScript UI
│   └── src/
│       ├── api/          # Typed Axios client
│       ├── components/   # Navbar, ImageUploader, ResultCard, ExplanationPanel, TrainingPanel
│       └── pages/        # HomePage, AnalyzePage, TrainPage
├── notebook/             # Research notebook (LeafExpert.ipynb)
├── docs/                 # Documentation
│   ├── README_classifier.md
│   ├── api_reference.md
│   └── setup_guide.md
├── assets/               # Sample images
├── docker-compose.yml
└── README.md

---

TODO

• Image Classifier (No-Code Interface)
• Explainer (Grad-CAM + LIME)
• VLM integration (planned)
• Integrating the Flow (FastAPI backend)
• React Frontend
• Docker deployment
• VLM captioning (Gemini / OpenAI Vision)
• Live demo deployment

---

License

MIT © Purushothaman Natarajan