Train custom computer vision defect detection model using Amazon SageMaker

This article provides a migration guide from Amazon Lookout for Vision (shutting down October 31, 2025) to Amazon SageMaker AI for computer vision defect detection models.

• AWS released pre-trained defect detection models on AWS Marketplace for fine-tuning in SageMaker
• Supports binary classification and semantic segmentation model types
• Offers greater flexibility: custom hyperparameters, adjustable training time, edge deployment options
• Step-by-step process: label data with Ground Truth, train model, deploy via endpoints or batch transform
• Use SageMaker Ground Truth or partner tools (Edge Impulse, Roboflow, SuperbAI) for data labeling
• Real-time inference via SageMaker endpoints; batch inference via Batch Transform for cost efficiency
• GitHub repository includes Jupyter Notebook for training and packaging models

This migration enables customers to continue using AWS defect detection technology with enhanced flexibility and control over model training and deployment.