Smart Classroom Attention Analyzer

An AI-powered web application that detects 12 student behaviour labels from classroom images using a pre-trained EfficientNet-B0 model.

🎯 Detected Behaviours

Emoji	Class	Description
📱	Using_phone	Student visibly using a mobile phone
🔄	bend	Bending posture
📚	book	Holding/looking at a book
🙇	bow_head	Head bowed down
🙋	hand-raising	Raising hand to participate
📞	phone	Phone (detected object)
🆙	raise_head	Head raised upward
📖	reading	Reading posture
😴	sleep	Sleeping in class
↩️	turn_head	Turning head sideways
🧍	upright	Upright, attentive posture
✍️	writing	Writing activity

🚀 Quick Start

Prerequisites

• Python 3.9 or later
• best_model.pth placed in the project root

1. Set up the virtual environment (Windows)

setup_env.bat

2. Activate the environment

venv\Scripts\activate

3. Run the web app

python app.py

4. Open your browser

Navigate to http://localhost:5000

🏗️ Project Structure

D_project/
├── app.py                  ← Flask web application
├── best_model.pth          ← Pre-trained model weights
├── requirements.txt        ← Python dependencies
├── setup_env.bat           ← Windows setup script
├── src/
│   ├── __init__.py
│   ├── model.py            ← EfficientNet-B0 architecture
│   └── predict.py          ← Inference pipeline
├── templates/
│   ├── index.html          ← Upload + results page
│   └── about.html          ← About page
└── static/
    ├── style.css           ← Dark glassmorphism UI
    └── script.js           ← Drag-and-drop + AJAX logic

🧠 Model Details

Property	Value
Architecture	EfficientNet-B0
Type	Multi-label Classification
Classes	12
Input Size	224 × 224 px
Detection Threshold	0.50 (Sigmoid)
Training Platform	Google Colab (Tesla T4 GPU)

📦 Tech Stack

• Backend: Python 3 · Flask
• Model: PyTorch · torchvision
• Frontend: Vanilla HTML/CSS/JS (dark glassmorphism design)

⚠️ Notes

• The best_model.pth file is included in this repo (≈17 MB, within GitHub's 100 MB limit).
• The dataset zip file (Student Behaviour Detection.v6i.multiclass.zip, 269 MB) is not included — it is listed in .gitignore.
• Multiple behaviours can be detected simultaneously in a single image (multi-label task).

📄 License

MIT License — free for educational and research use.