get_train_data.py

from datasets import dataset_utils
import tensorflow as tf


slim = tf.contrib.slim


_NUM_CLASSES = 9
_SPLITS_TO_SIZES = {'train': 10868, 'test': 4000}


_ITEMS_TO_DESCRIPTIONS = {
    'image': 'A [299 x 299 x 1] grayscale image.',
    'label': 'A single integer between 0 and 8',
}

def get_dataset(name, split_name, dataset_dir):
	reader = tf.TFRecordReader
	keys_to_features = {
      'image/encoded': tf.FixedLenFeature((), tf.string, default_value=''),
      'image/format': tf.FixedLenFeature((), tf.string, default_value='png'),
      'image/class/label': tf.FixedLenFeature(
          [], tf.int64, default_value=tf.zeros([], dtype=tf.int64)),
    }

        items_to_handlers = {
         'image': slim.tfexample_decoder.Image(shape=[299, 299, 1], channels=1),
         'label': slim.tfexample_decoder.Tensor('image/class/label', shape=[]),
    }

        decoder = slim.tfexample_decoder.TFExampleDecoder(
         keys_to_features, items_to_handlers)


        labels_to_names = None
        dataset_dir = dataset_dir+'/'+name+'.tfrecords'
        return slim.dataset.Dataset(
             data_sources=dataset_dir,
             reader=reader,
             decoder=decoder,
             num_samples=_SPLITS_TO_SIZES[split_name],
             num_classes=_NUM_CLASSES,
             items_to_descriptions=_ITEMS_TO_DESCRIPTIONS,
             labels_to_names=labels_to_names)