GCP Professional Cloud Architect Practice Question

Creating a Vertex AI Managed Dataset that simply references the Cloud Storage object URIs gives the pipeline a first-class, versionable resource that stores metadata and annotations while leaving the image files in place. Vertex AI datasets integrate directly with Data Labeling so new items can be queued for human labeling, and the same dataset can be reused by any subsequent AutoML or custom-training step without duplicating data. Passing raw bucket paths, storing lists in Artifact Registry, or using gcsfuse + Git lacks tight integration with Vertex AI labeling and lineage tracking. Importing images into BigQuery and training with AutoML Tables is inappropriate because Tables expects structured tabular data, not images.