TerraSight is preparing to train the SageMaker built-in Object Detection algorithm on images annotated by Ground Truth. The Ground Truth output is an augmented manifest (JSON Lines) that references image S3 URIs and bounding-box coordinates. Which data format should TerraSight use to import the labeled bounding-box dataset to the SageMaker built-in Object Detection training job?

The SageMaker built-in Object Detection algorithm directly supports the Augmented Manifest file format generated by Amazon Ground Truth. This JSON Lines format, referencing S3 URIs for images and bounding box annotations, is a native input type for the algorithm. Therefore, TerraSight can use the Ground Truth output as-is. Incorrect options: The SageMaker built-in Object Detection algorithm supports multiple formats, including Augmented Manifest, not exclusively COCO. While COCO is a valid format, it's not the only one, and conversion is unnecessary here. MXNet RecordIO is another supported format, but the algorithm also accepts Augmented Manifests, making conversion to RecordIO unnecessary and an extra step. The SageMaker built-in Object Detection algorithm does not natively read annotations from relational databases via JDBC. Its input channels are designed for file-based data stored in S3.