QuestionQ86

VerifiedAutomating and orchestrating ML pipelines

You work on the AI team at an automobile company and are building a visual defect-detection model with TensorFlow and Keras. To improve model performance, you want to add image-augmentation functions such as translation, cropping, and contrast adjustment. These functions are randomly applied to every training batch. You want to optimize the data-processing pipeline for runtime and compute-resource utilization. What should you do?

Explanation

Dynamic augmentation in a tf.data pipeline keeps transformations in TensorFlow’s optimized input pipeline. Dataset.map can parallelize independent preprocessing work, and prefetching can overlap preprocessing with model execution. Generating augmentations on demand also avoids storing the large set of possible augmented-image TFRecords.

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!