A mining company wants to use machine learning (ML) models to identify mineral images in real time. A data science team built an image recognition model that is based on convolutional neural network…

This choice is an incomplete description of the full SageMaker deployment process, missing the crucial step of using SageMaker's Endpoint Configuration to define instance types.

Amazon SageMaker Model Registry is used to manage and version models, facilitating deployment. From the Model Registry, the model can be directly deployed to a SageMaker Endpoint, where the Endpoint Configuration allows specifying the instance type (e.g., GPU instances for CNN inference) and other parameters based on known workload patterns. This integrated SageMaker workflow offers the most streamlined, least development-intensive approach for deploying models for real-time inference.

While deploying to GPU instances is appropriate, explicitly involving AWS Lambda and Amazon API Gateway adds unnecessary development effort because SageMaker endpoints inherently provide HTTPS inference and scaling capabilities, making a custom wrapper redundant.

Deploying to CPU instances for a CNN-based image recognition model, especially one trained on GPUs, would likely result in unacceptably slow real-time inference, failing to meet performance expectations for identifying mineral images in real-time.