Microsoft Certified: Azure AI Engineer AssociateImplement image and video processing solutionsHard
A security firm is developing an automated surveillance system for a remote industrial site with limited internet connectivity. The system needs to detect unauthorized personnel and trigger an alarm in real-time, even if the connection to Azure is temporarily lost. Which deployment option for an Azure Custom Vision model is most appropriate?
- AUse Azure Machine Learning for batch inference.
- BDeploy as a cloud endpoint for real-time inference.
- CExport the model for deployment to an IoT Edge device.
- DIntegrate with Azure Video Indexer for person detection.
Show answer & explanationAnswer & explanation
Correct answer: C. Export the model for deployment to an IoT Edge device.
Exporting an Azure Custom Vision model for deployment to an IoT Edge device allows the model to run locally on the device (at the 'edge'). This enables real-time inference and alarm triggering even with intermittent or lost internet connectivity, fulfilling the requirement for a remote site.
Why the other options are wrong
- A. Azure Machine Learning batch inference is for processing large datasets offline, not for real-time detection on a live video stream at the edge.
- B. A cloud endpoint requires continuous internet connectivity for inference, which is not guaranteed at a remote site.
- D. Video Indexer is a cloud service for comprehensive video analysis; it doesn't provide edge deployment for real-time, offline inferencing.
Custom Vision IoT Edge Deployment
The ability to export a trained Azure Custom Vision model to run locally on an IoT Edge device, enabling real-time inference at the network edge.
- Enables offline inference
- Reduces latency for real-time applications
- Conserves bandwidth by processing data locally
- Supports various export formats (ONNX, Dockerfile, TensorFlow, etc.)
Memory trick: Edge Devices Run Custom Models Offline.