A social media company will train an image-classification model (using SageMaker’s built-in image classifier) to detect offensive content. They will use pipe mode to speed training. The dataset is split into Training, Validation, and Test folders, with class subfolders, images resized uniformly, and two manifest files named training.lst and validation.lst. They created separate S3 buckets for training and validation uploads. What additional data-preparation step should they perform before uploading to S3?

The correct answer is to convert the manifests into RecordIO files using the im2rec utility. SageMaker's built-in image classification algorithm, when using pipe mode, expects data in RecordIO format for optimal performance. The im2rec utility from Apache MXNet is specifically designed to convert image datasets and their corresponding manifest files into the RecordIO format, which is highly optimized for sequential data access and efficient I/O during training. Incorrect options: Pandas DataFrames and Parquet files are not the expected input format for the SageMaker built-in image classifier in pipe mode. Compressing directories with Snappy or gzip is not the correct approach; the algorithm expects RecordIO files, not compressed directories of individual images. While compression can save space, it doesn't provide the optimized data access that RecordIO offers for this specific algorithm and mode.