AnomaVision: Edge-Ready Visual Anomaly Detection

Lightweight, fast, and production-ready anomaly detection powered by PaDiM. Deploy anywhere: edge devices, servers, or the cloud.

---

Overview

AnomaVision delivers state-of-the-art visual anomaly detection optimized for real-world deployment. Built for speed and efficiency, it outperforms existing solutions while maintaining a small footprint perfect for edge devices.

Key Features

• 🎯 Superior Performance — Higher AUROC across MVTec AD and Visa datasets
• ⚡ 3× Faster Inference — Optimized for both CPU and GPU deployment
• 📦 Smaller Models — 30MB models with lower memory footprint
• 🌐 Multi-Backend Export — PyTorch, ONNX, TorchScript, OpenVINO, TensorRT, INT8 Quantization
• 🖥️ Production Ready — Python API, CLI tools, C++ runtime, and REST API
• 🎨 Rich Visualizations — Heatmaps, bounding boxes, and ROC curves
• 🎮 Interactive Demo — Streamlit web interface for instant testing

---

Why Choose AnomaVision?

Performance Advantages Over Anomalib

CPU Inference:

Metric	AnomaVision	Anomalib	Improvement
Training Time (s)	8.38	13.07	-35.9%
Inference FPS	43.41	13.03	+233%
ms / image	23.0	76.7	-70%

GPU Inference (CUDA):

Metric	AnomaVision	Anomalib	Improvement
Training Time (s)	8.38	13.07	-35.9%
Inference FPS	547.46	355.72	+53.9%
ms / image	1.83	2.81	-35.0%

Accuracy:

• MVTec AD: Image AUROC 0.85 vs 0.81 | Pixel AUROC 0.96 vs 0.94
• Visa: Image AUROC 0.81 vs 0.78 | Pixel AUROC 0.96 vs 0.95

💡 Download Full Performance Analysis (PDF) | Detailed Benchmarks

---

Quick Start

Installation

Using Poetry (Recommended):

git clone https://github.com/DeepKnowledge1/AnomaVision.git
cd AnomaVision
poetry install
poetry shell

Using pip:

pip install AnomaVision

📖 Detailed Installation Guide

Basic Usage

Training:

python train.py --config config.yml
# Outputs: padim_model.pt, padim_model.pth, config.yml

Detection:

python detect.py --config config.yml

Evaluation:

python eval.py --config config.yml

Export:

python export.py --config export_config.yml

📖 Complete Quick Start Guide

---

Interactive Demo

Real-time anomaly detection with explainable heatmaps

Experience AnomaVision through our intuitive web interface:

• ⚡ Real-time anomaly detection
• 🎯 Explainable AI with visual heatmaps
• 📊 Interactive threshold controls
• 🔍 Batch processing support
• 💾 Export results as JSON
• 🎮 No coding required

Launch the Demo

# Start FastAPI backend
uvicorn apps.api.fastapi_app:app --host 0.0.0.0 --port 8000

# Launch Streamlit demo (new terminal)
streamlit run apps/ui/streamlit_app.py -- --port 8000

Open http://localhost:8501 in your browser.

📖 Streamlit Demo Guide

---

Deployment Options

Method	Best For	Key Benefits
🎨 Streamlit Demo	Testing, demonstrations	Zero-code UI, instant feedback
📌 FastAPI Backend	Production APIs	REST endpoints, scalable
🖥️ C++ Runtime	Edge devices	No Python dependency, ultra-fast
💻 Python CLI	Batch processing	Scriptable, configurable
📦 PyPI Package	Custom integration	Import as library

REST API Example

import requests

with open("test_image.jpg", "rb") as f:
    response = requests.post(
        "http://localhost:8000/predict",
        files={"file": f},
        params={"include_visualizations": True}
    )
    result = response.json()
    print(f"Anomaly Score: {result['anomaly_score']}")
    print(f"Is Anomaly: {result['is_anomaly']}")

📖 FastAPI Setup Guide

C++ Inference

Deploy without Python using our ONNX Runtime + OpenCV implementation:

• 🖼️ Complete pipeline from preprocessing to visualization
• 📦 Modular architecture
• 🌐 Perfect for edge devices

📖 C++ Inference Guide

---

Use Cases

• 🏭 Manufacturing QC — Real-time defect detection on production lines
• 🔬 Medical Imaging — Anomaly identification in X-rays, MRIs, microscopy
• 🏗️ Infrastructure — Crack and corrosion detection
• 📱 PCB Inspection — Soldering defects and component issues
• 🌾 Agriculture — Plant disease and crop monitoring
• 🚗 Automotive — Paint defects and assembly quality

---

Documentation

• 📖 Installation
• 🚀 Quick Start
• 🎨 Streamlit Demo
• 📌 FastAPI Backend
• 💻 CLI Reference
• 📚 API Reference
• ⚙️ Configuration
• 📊 Benchmarks
• 🔧 Troubleshooting
• 🤝 Contributing

---

Community & Support

• 💬 GitHub Discussions
• 🐛 Issue Tracker
• 📧 Email Support

---

Citation

@software{anomavision2025,
  title={AnomaVision: Edge-Ready Visual Anomaly Detection},
  author={DeepKnowledge Contributors},
  year={2025},
  url={https://github.com/DeepKnowledge1/AnomaVision},
}

---

Acknowledgments

Built on the foundation of Anodet. We thank the original authors for their contributions to open-source anomaly detection research.

---

License

Released under the MIT License.

---

Ready to get started? Follow our Quick Start Guide and build your first anomaly detection pipeline in 5 minutes!