nerdexam
Google

PROFESSIONAL-MACHINE-LEARNING-ENGINEER · Question #260

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 mod

The correct answer is C. Convert the images to TFRecords and store them in a Cloud Storage bucket. Read the. To efficiently read over one million images at scale for TensorFlow model training on Vertex AI while minimizing I/O bottlenecks, convert the images to TFRecords and store them in Cloud Storage.

Submitted by hans_de· Apr 18, 2026Data processing and feature engineering

Question

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?

Options

  • ALoad the images directly into the Vertex AI compute nodes by using Cloud Storage FUSE. Read
  • BCreate a Vertex AI managed dataset from your image data. Access the
  • CConvert the images to TFRecords and store them in a Cloud Storage bucket. Read the
  • DStore the URLs of the images in a CSV file. Read the file by using the

How the community answered

(32 responses)
  • A
    6% (2)
  • B
    9% (3)
  • C
    81% (26)
  • D
    3% (1)

Why each option

To efficiently read over one million images at scale for TensorFlow model training on Vertex AI while minimizing I/O bottlenecks, convert the images to TFRecords and store them in Cloud Storage.

ALoad the images directly into the Vertex AI compute nodes by using Cloud Storage FUSE. Read

While Cloud Storage FUSE allows mounting a bucket, it can introduce overhead and might not be as optimized for high-throughput, random access reading of millions of small files as TFRecords, potentially leading to I/O bottlenecks.

BCreate a Vertex AI managed dataset from your image data. Access the

Creating a Vertex AI managed dataset can simplify data management, but it doesn't inherently optimize the I/O performance for reading raw image files at scale for TensorFlow training; the underlying format still matters.

CConvert the images to TFRecords and store them in a Cloud Storage bucket. Read theCorrect

Converting images to TFRecords (or other similar binary formats) allows for efficient serialization of data, grouping multiple examples into larger files, and optimizing I/O operations. TensorFlow's `tf.data.TFRecordDataset` is highly optimized for reading these files in parallel, minimizing bottlenecks when streaming large image datasets from Cloud Storage during training.

DStore the URLs of the images in a CSV file. Read the file by using the

Storing URLs in a CSV and reading images one by one directly from Cloud Storage is inefficient due to many small HTTP requests and network latency, which will significantly bottleneck training for a large dataset.

Concept tested: Efficient large-scale image data loading for TensorFlow training

Source: https://www.tensorflow.org/tutorials/load_data/tfrecord

Topics

#TFRecords#Data I/O optimization#TensorFlow data pipelines#Cloud Storage

Community Discussion

No community discussion yet for this question.

Sign up to join the discussion
Full PROFESSIONAL-MACHINE-LEARNING-ENGINEER Practice