Anomaly Detection API

🚀 Overview

FastAPI-based anomaly detection system for video frames, running completely locally with no cloud dependencies. Originally developed on Azure Databricks, now fully independent for cost-free operation.

📊 Prediction Results

🎥 Demo Video

🎬 Watch Demo Video (7.5MB)

Live demonstration of the Streamlit web interface with drag & drop image upload and real-time anomaly detection results

Note: Video is hosted in GitHub Releases

Interactive Demo

🔗 View Live Results (GitHub Pages)

Jupyter Notebook

📓 Interactive Testing Notebook - Run this notebook to see live predictions with HTML visualization

Real-time anomaly detection results showing normal frames (green border) and anomalous frames (red border) with confidence scores.

🛠️ Quick Start

1. Setup Environment

# Create virtual environment
python -m venv .venv
source .venv/bin/activate  # On Windows: .venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

2. Download Model (One-time setup)

# Download model from Databricks to local storage
# This creates models/model/ directory with the MLflow model
python download_model.py

3. Start API Server

# Use local model (Databricks-independent)
python api_local.py

Note: api.py is no longer available as the Azure Databricks workspace has been deleted. Only api_local.py works with the downloaded local model.

4. Test with Client

python client.py

5. Interactive Testing

Option A: Jupyter Notebook

Open notebook/07_deployment.ipynb for interactive testing with HTML visualization.

Option B: Web Interface (Upload Your Own Images) 🆕

# Install Streamlit (if not already installed)
pip install streamlit

# Start the local API server first
python api_local.py

# In another terminal, run the web interface
streamlit run web_interface.py

Then open your browser to http://localhost:8501 and upload your own images for real-time predictions!

📁 Project Structure

anomaly_detection/
├── api.py                 # FastAPI server
├── client.py              # API client for testing
├── web_interface.py       # Streamlit web interface for image upload
├── frames/                # Sample images
├── notebook/              # Jupyter notebooks
│   └── 07_deployment.ipynb # Interactive testing & visualization
└── requirements.txt       # Dependencies

🔧 Features

• 🏠 Fully Local: No cloud dependencies or costs
• 📤 Base64 Image Processing: Handles multiple images in batch
• ⚡ Real-time Predictions: Fast inference with CPU optimization
• 🎨 Visual Results: Color-coded prediction display
• 🌐 Web Interface: Drag & drop image upload with Streamlit
• 🛡️ Error Handling: Comprehensive logging and fallback mechanisms
• 💰 Cost-Free: Zero ongoing operational costs

🎯 API Endpoints

• GET /: Health check
• POST /predict: Batch image prediction

📝 Model Details

• Framework: Transformers pipeline via MLflow
• Input: Base64-encoded images
• Output: Classification scores (normal/anomaly)
• Deployment: Local FastAPI (Databricks alternative)