Synthetic data for the safety cases that are hardest to photograph

Industrial safety AI needs training images of people in dangerous proximity, but those situations are scarce in training data and hazardous to photograph. AWS describes a SageMaker AI and Rekognition pipeline that generates photorealistic training images with automated labels. In its own experiments, AWS reports up to a 160% improvement in person-detection mAP50, without manual annotation or hazardous photography sessions. The practical lesson is specific: synthetic data can fill a hard-to-collect edge case. The excerpt does not establish how the models perform in live workplaces.