You are developing an ML model to identify your company’s products in images. You have access to over one million images in a Cloud Storage bucket. You plan to experiment with different TensorFlow models by using Vertex AI Training. You need to read images at scale during training while minimizing data I/O bottlenecks. What should you do?