CIFAR10 AutoML

cifar10_automl.py

import tensorflow as tf
import tensorflow_datasets as tfds

cifar10 = tfds.load('cifar10', as_supervised=True)
class_names = ['airplane', 'automobile', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck']

def save_images_write_metadata(dataset, purpose, mf):
  i = 1
  for image, label in dataset:
    fn = '{}_{}.jpeg'.format(purpose, i)
    tf.keras.preprocessing.image.save_img(TEMP_LOCAL_PATH + fn, image.numpy())
    metadata = {
      'imageGcsUri': GCS_PATH_FOR_DATA + fn,
      'classificationAnnotation': {'displayName': class_names[label.numpy()]},
      'dataItemResourceLabels': {'aiplatform.googleapis.com/ml_use': purpose}
    }
    json.dump(metadata, mf)
    mf.write('\n')
    i += 1

with open('metadata.jsonl', 'w') as mf:
  save_images_write_metadata(cifar10['train'].take(40000), 'train', mf)
  save_images_write_metadata(cifar10['train'].skip(40000), 'validation', mf)
  save_images_write_metadata(cifar10['test'], 'test', mf)