🥭 Mango Disease AI — Documentation

Step 1 — Get Started

Install in 30 Seconds

Install from PyPI — the official Python package index. Works on Windows, Mac, Linux.

Open PowerShell / Terminal

Press Windows Key → type PowerShell → press Enter

Install the package

Copy and paste the command on the right → press Enter

Wait for installation

Takes ~30 seconds. You'll see: Successfully installed mango-disease-ai-0.1.1

Install API server tools

Run the second command to install FastAPI server tools

First run note ⚠️

The CLIP model (~600 MB) downloads automatically once. Takes 1-2 min. All future runs are instant.

PowerShell

# Install the PyPI library

PS> pip install mango-disease-ai

✓ Successfully installed mango-disease-ai-0.1.1

# Install REST API server tools

PS> pip install fastapi "uvicorn[standard]" python-multipart

✓ All API dependencies installed

# Verify installation

PS> python -c "import mango_disease_ai; print(mango_disease_ai.__version__)"

0.1.1

⚡ For Anaconda users (Windows)

PS> & "E:\Anaconda\python.exe" -m pip install mango-disease-ai

Section A — Python Developers

PyPI Library — Python Usage

Import directly in your Python code. Just 2 functions: analyze() and generate_pdf()

Basic Usage

Full Output

PDF Report

Fast Mode

my_script.py

from mango_disease_ai import analyze

# Analyze any mango image

result = analyze("path/to/mango_leaf.jpg")

# Check if it's a mango

if result["is_mango"]:

print(result["predicted_class"])

print(result["confidence"])

📤 Console Output

# First run (downloads CLIP ~600MB once)

Loading weights: 100%|████| 398/398

Healthy

0.9470

🎯 Returns disease name + confidence score (0.0 to 1.0)

full_analysis.py

from mango_disease_ai import analyze

result = analyze("mango.jpg")

if result["is_mango"]:

# All 7 class scores

for s in result["all_scores"]:

print(s['class'], s['score'])

# Disease info

info = result["disease_info"]

print(info["scientific_name"])

print(info["symptoms"])

print(info["remedies"])

📊 Full Output Preview

🥭 Disease: Healthy (94.7%)

All 7 class scores:

Scientific name:

Mangifera indica (Normal)

Symptoms:

• Vibrant green leaf coloration
• Smooth and glossy leaf surface
• No spots or lesions

generate_report.py

from mango_disease_ai import analyze, generate_pdf

import base64

result = analyze("mango.jpg")

if result["is_mango"]:

# Save heatmap PNG

heatmap = base64.b64decode(result["gradcam_base64"])

with open("heatmap.png", "wb") as f:

f.write(heatmap)

# Generate PDF

pdf = generate_pdf(result, user_name="Dr. Rahman")

with open("report.pdf", "wb") as f:

f.write(pdf)

📄 Output Files Generated

🌡️ heatmap.png

Grad-CAM++ overlay showing where the AI focused. Red = high attention areas. ~50-100 KB PNG at 224×224px.

📋 report.pdf

Professional PDF with your name, date, disease name, all scores, original + heatmap images, symptoms & treatment. ~180 KB.

fast_mode.py

from mango_disease_ai import analyze

# Fast mode — skip heatmap generation

# ~2 seconds vs ~8 seconds with heatmap

result = analyze("mango.jpg", include_gradcam=False)

print(result["predicted_class"]) # works!

print(result["gradcam_base64"]) # None

⚡ Speed Comparison

Fast mode (no heatmap)~2 sec

Full mode (with heatmap)~8 sec

💡 Use fast mode when you only need the disease name and don't need the heatmap visualization

Section B — All Developers

REST API — HTTP Endpoints

Use the API from any language: Python, JavaScript, Java, Kotlin, Swift, PHP — anything that can send HTTP requests.

Start the API Server

# Step 1: Go to the project folder

PS> cd "E:\AIUB R&D ICCA\Fully Final Mango api , without api\Mango API"

# Step 2: Start the server

PS> & "E:\Anaconda\python.exe" start_api.py

INFO: Application startup complete.

INFO: Uvicorn running on http://0.0.0.0:8000

# Step 3: Open browser → http://localhost:8000/docs

🌐 Browser (Swagger)

🐍 Python

⚡ JavaScript

🔧 curl

Testing in the Browser (Swagger UI)

Open: http://localhost:8000/docs

You see all 5 endpoints listed with green (GET) and yellow (POST) badges

Click any endpoint to expand it

E.g. click POST /api/analyze — it opens a full form

Click "Try it out"

This activates the form so you can fill it in

Upload your image + click "Execute"

Click "Choose File" → select your mango image → click the blue Execute button

See the live JSON response

Scroll down — you see the full JSON with disease name, confidence, heatmap, everything

api_test.py

import requests

# ── 1. Health Check ──────────────────────

r = requests.get("http://localhost:8000/api/health")

print(r.json()) # {'status': 'ok', 'version': '0.1.1'}

# ── 2. Analyze Image ─────────────────────

with open("mango.jpg", "rb") as f:

r = requests.post(

"http://localhost:8000/api/analyze",

files={"image": ("mango.jpg", f, "image/jpeg")},

data={"include_gradcam": "false"},

)

result = r.json()

print("Disease:", result["predicted_class"])

print("Confidence:", f"{result['confidence']:.1%}")

# ── 3. Download PDF ──────────────────────

with open("mango.jpg", "rb") as f:

r = requests.post("http://localhost:8000/api/report",

files={"image": f}, data={"user_name": "Dr. Rahman"})

open("report.pdf", "wb").write(r.content)

app.js

async function analyzeMango(imageFile) {

const formData = new FormData();

formData.append("image", imageFile);

formData.append("include_gradcam", "false");

const response = await fetch(

"http://localhost:8000/api/analyze",

{ method: "POST", body: formData }

);

// Handle non-mango images (HTTP 422)

if (response.status === 422) {

alert("Not a mango! Please upload a mango image.");

return;

}

const result = await response.json();

console.log("Disease:", result.predicted_class);

console.log("Confidence:", (result.confidence * 100).toFixed(1) + "%");

// Display heatmap in <img> tag

document.getElementById("heatmap").src =

"data:image/png;base64," + result.gradcam_base64;

}

Terminal

# Health check

$ curl http://localhost:8000/api/health

{"status":"ok","version":"0.1.1"}

# Analyze image

$ curl -X POST http://localhost:8000/api/analyze \

-F "image=@mango.jpg" \

-F "include_gradcam=false"

# Download PDF

$ curl -X POST http://localhost:8000/api/report \

-F "image=@mango.jpg" \

-F "user_name=Dr. Rahman" \

--output report.pdf

All Endpoints

5 API Endpoints

Every endpoint, what it does, and exactly what it returns.

GET /api/health ~10ms

Check if the server is alive and running. Use this before sending image requests.

Response: {"status": "ok", "version": "0.1.1"}

GET /api/diseases ~5ms

Get full information about all 7 detectable diseases: scientific name, description, symptoms, and remedies.

{

"diseases": {

"Anthracnose": {

"scientific_name": "Colletotrichum gloeosporioides",

"symptoms": ["Dark brown spots", "..."],

"remedies": ["Apply copper fungicide", "..."]

"Bacterial Canker": { ... },

// ... all 7 diseases

}

POST /api/analyze ~2-8 sec

Upload a mango image → get full disease diagnosis. Body: image (file) + include_gradcam (bool). Returns 422 if not a mango image.

{

"is_mango": true,

"mango_confidence": 0.82,

"predicted_class": "Healthy",

"confidence": 0.9470,

"all_scores": [

{ "class": "Healthy", "score": 0.947 },

{ "class": "Anthracnose", "score": 0.012 }, ...

"disease_info": { "scientific_name": "...", "symptoms": [...] },

"gradcam_base64": "iVBORw0KGgoAAAANSUhEUg..." // PNG image

}

POST /api/report ~10-15 sec

Upload image + your name → download a professional PDF report directly. Body: image (file) + user_name (string).

Response: application/pdf binary file (~180 KB)

Header: Content-Disposition: attachment; filename="mango_diagnosis_Dr_Rahman.pdf"

GET /docs Swagger UI

Interactive Swagger UI — test all endpoints directly in the browser. Auto-generated from the code.

Open Swagger UI ↗

How It Works

System Architecture

The AI pipeline that runs every time you analyze a mango image.

📸 Input: Image (JPG/PNG/WebP/BMP)

Any mango leaf or fruit photo

↓

🔍 CLIP Validation

Checks: "Is this a mango?"

❌ Not mango
→ HTTP 422

↓ ✅ Is mango

🧠 AA-ENet Classification

EfficientNet-B0 + CBAM + Transformer

↓

📊 Disease + Scores

7 class probabilities

🌡️ Grad-CAM++

Heatmap visualization

↓

✅ Result JSON / PDF Report

Disease + Confidence + Heatmap + Info

🧠 AA-ENet Model

Custom EfficientNet-B0 backbone with CBAM attention and Transformer fusion. Trained on Amropali mango dataset.

EfficientNet-B0CBAMTransformer

🔍 CLIP Validation

OpenAI CLIP (ViT-B/32) validates that the uploaded image is actually a mango — prevents false diagnoses on random photos.

openai/clip-vit-base-patch32Zero-shot

🌡️ Grad-CAM++ Heatmap

Shows exactly where the AI looked to make its decision. Red = high focus, Blue = ignored. Useful for research validation.

Grad-CAM++CBAM layer

📄 PDF Report Engine

fpdf2-based report generator creates professional diagnosis reports with images, scores, symptoms & treatment.

fpdf2180KB PDF

Project Files

File Structure

Mango API/

📁 Mango API/

├── run_mango.py ← Test the PyPI library

├── start_api.py ← One-click API launcher

├── test_api.py ← Full 6-test suite

├── single_test.py ← Test ONE endpoint

├── USAGE_GUIDE.md ← Full documentation

├── Guide For install.txt ← Quick start

└── 📁 api/

├── main.py ← FastAPI application

├── schemas.py ← Pydantic models

├── __init__.py ← Package init

└── requirements.txt ← API deps

Technology Stack

🔬 AI/ML

PyTorchtimmCLIPOpenCV

🌐 API

FastAPIUvicornPydantic

📦 Package

PyPIsetuptoolstwine

📄 Report

fpdf2PillowNumPy

Cheat Sheet

All Commands

Quick Reference — Copy Any Command

# ── Install ──────────────────────────────────────────

PS> & "E:\Anaconda\python.exe" -m pip install mango-disease-ai

PS> & "E:\Anaconda\python.exe" -m pip install fastapi "uvicorn[standard]" python-multipart

# ── Navigate to project ──────────────────────────────

PS> cd "E:\AIUB R&D ICCA\Fully Final Mango api , without api\Mango API"

# ── Test PyPI library ────────────────────────────────

PS> & "E:\Anaconda\python.exe" run_mango.py

# ── Start REST API server ────────────────────────────

PS> & "E:\Anaconda\python.exe" start_api.py

# ── Run ALL API tests (6 tests) ──────────────────────

PS> & "E:\Anaconda\python.exe" test_api.py

# ── Test ONE specific endpoint ───────────────────────

PS> & "E:\Anaconda\python.exe" single_test.py health

PS> & "E:\Anaconda\python.exe" single_test.py analyze

PS> & "E:\Anaconda\python.exe" single_test.py report

# ── Open Swagger UI in browser ───────────────────────

PS> start http://localhost:8000/docs

# ── Check installed version ──────────────────────────

PS> & "E:\Anaconda\python.exe" -c "import mango_disease_ai; print(mango_disease_ai.__version__)"

🥭 Mango Disease
AI — Complete Docs

Install in 30 Seconds

Open PowerShell / Terminal

Install the package

Wait for installation

Install API server tools

First run note ⚠️

PyPI Library — Python Usage

REST API — HTTP Endpoints

Testing in the Browser (Swagger UI)

Open: http://localhost:8000/docs

Click any endpoint to expand it

Click "Try it out"

Upload your image + click "Execute"

See the live JSON response

5 API Endpoints

Predicted Output Types

All 7 Disease Classes

System Architecture

File Structure

Technology Stack

All Commands

🥭 Mango DiseaseAI — Complete Docs

Install in 30 Seconds

Open PowerShell / Terminal

Install the package

Wait for installation

Install API server tools

First run note ⚠️

PyPI Library — Python Usage

REST API — HTTP Endpoints

Testing in the Browser (Swagger UI)

Open: http://localhost:8000/docs

Click any endpoint to expand it

Click "Try it out"

Upload your image + click "Execute"

See the live JSON response

5 API Endpoints

Predicted Output Types

All 7 Disease Classes

System Architecture

File Structure

Technology Stack

All Commands

🥭 Mango Disease
AI — Complete Docs