A practical guide for real-world and efficient computer vision applications for resource-constrained devices with industry standards in mind.
- Start with the Edge AI Engineering: a practical guide covering core concepts of the entire Edge AI MLOps stack with industry blueprints.
- Then read this: The Next AI Frontier is at the Edge
- Related work: Edge Language | Edge Audio
The goal of this guide is to provide resources for building, optimizing, and deploying Computer Vision applications at the edge, through hands-on examples including practical notebooks and real-world use cases across key industries.
Industry Blueprints
- Autonomous Systems
- Healthcare & Medical Imaging*
- Retail & Consumer Analytics
- Security & Surveillance
- Agriculture & Precision Farming
- Manufacturing & Quality Control
- Smart Cities & Urban Planning
Edge Optimization Lab: techniques and tools for maximizing performance and efficiency of vision models on edge hardware
- Model Quantization
- Pruning Techniques
- Federated Learning
- Compiler Targets
- Hardware-Specific Optimization
Production Pipelines: guides and templates for robust, scalable edge vision AI operations
- CI/CD for Edge
- Monitoring (Drift Detection, Edge Metrics Dashboard)
- OTA Updates
- Edge Security (Secure Boot, Data Encryption, Threat Detection, Privacy-Preserving vision, Adversarial Robustness, Device Hardening, Compliance)
Reference Architectures: blueprints for edge vision hardware and system design
- Microphone Array Setups
- Edge Server Specs
- IoT Connectivity
- Edge-Cloud Hybrid Models
Integration
- Notebooks (hands-on deep dives)
- Companion Resources
- Industry-Specific Stardards
├── edge-ai-engineering/
│ ├── introduction-to-edge-ai.md
│ ├── edge-ai-architectures.md
│ ├── model-optimization-techniques.md
│ ├── hardware-acceleration.md
│ ├── edge-deployment-strategies.md
│ ├── real-time-processing.md
│ ├── privacy-and-security.md
│ ├── edge-ai-frameworks.md
│ └── benchmarking-and-performance.md
├── industry-blueprints/
│ ├── autonomous-systems/
│ │ ├── traffic-analysis-yolov8-tensorrt.md
│ │ ├── drone-navigation-lite.md
│ │ ├── pedestrian-tracking-edgetpu.md
│ │ └── vehicle-defect-detection-openvino.md
│ ├── healthcare-medical-imaging/
│ │ ├── xray-classification-tflite.md
│ │ ├── ultrasound-segmentation-ncnn.md
│ │ ├── mri-tumor-detection-onnx.md
│ │ └── remote-patient-monitoring-jetson.md
│ ├── retail-consumer-analytics/
│ │ ├── shelf-analytics-mmdetection.md
│ │ ├── checkout-automation.md
│ │ ├── customer-behavior-analysis-openvino.md
│ │ └── inventory-management-edge-tflite.md
│ ├── security-surveillance/
│ │ ├── perimeter-surveillance-yolo.md
│ │ ├── anomaly-detection-autoencoder.md
│ │ ├── facial-recognition-privacy-preserving.md
│ │ └── crowd-behavior-analysis-edge.md
│ ├── agriculture-precision-farming/
│ │ ├── crop-health-monitoring-multispectral.md
│ │ ├── yield-prediction-edge-ml.md
│ │ └── autonomous-harvesting-robotics.md
│ ├── manufacturing-quality-control/
│ │ ├── defect-detection-openvino.md
│ │ ├── robotic-picking-ort.md
│ │ └── predictive-maintenance-edge-analytics.md
│ └── smart-cities-urban-planning/
│ ├── traffic-flow-optimization-edge.md
│ ├── waste-management-vision-ai.md
│ └── energy-grid-monitoring-federated.md
├── edge-optimization-lab/
│ ├── model-quantization/
│ │ ├── post-training-int8.md
│ │ └── qat-pytorch.md
│ ├── pruning-techniques/
│ │ ├── magnitude-pruning.md
│ │ └── lottery-ticket-hypothesis.md
│ ├── federated-learning/
│ │ ├── privacy-preserving-cv.md
│ │ └── distributed-training.md
│ ├── compiler-targets/
│ │ ├── tvm-tutorial.md
│ │ └── onnx-runtime-guide.md
│ └── hardware-specific-optimization/
│ ├── nvidia-jetson-optimization.md
│ ├── raspberry-pi-edge-ai.md
│ └── microcontroller-tinyml.md
├── production-pipelines/
│ ├── ci-cd-for-edge.md
│ ├── monitoring/
│ │ ├── drift-detection.md
│ │ └── edge-metrics-dashboard.md
│ ├── ota-updates.md
│ └── edge-security/
│ ├── secure-boot-implementation.md
│ ├── data-encryption-edge.md
│ ├── threat-detection/
│ │ ├── perimeter-surveillance.md
│ │ └── anomaly-detection.md
│ ├── privacy-preserving-cv/
│ │ ├── federated-learning-techniques.md
│ │ └── differential-privacy.md
│ ├── model-security/
│ │ └── adversarial-robustness.md
│ ├── edge-device-hardening/
│ │ ├── secure-deployment.md
│ │ └── secure-communication.md
│ └── industry-compliance/
│ ├── regulatory-standards.md
│ └── ethical-ai-guidelines.md
├── reference-architectures/
│ ├── industrial-camera-setups.md
│ ├── edge-server-specs.md
│ ├── iot-connectivity.md
│ └── edge-cloud-hybrid-models.md
└── _integration/
├── cs-notebook-redirects.md
├── companion-resources.md
└── industry-specific-regulations.md
Important
This project uses a submodule edge-ai-engineering located in lab/edge-ai-engineering.
Please initialize submodules after cloning the repository: git submodule update --init --recursive
- Clone this repository:
git clone https://github.com/afondiel/edge-vision.git- Explore the Edge AI Engineering for foundational knowledge.
- Dive into Industry Blueprints for hands-on, sector-specific language AI guides.
- Use the Edge Optimization Lab and Production Pipeline for deployment and scaling.
We welcome contributions! Please see our CONTRIBUTING.md file for details on how to submit improvements.
This project is licensed under the MIT License - see the LICENSE.md file for details.
Books: