main.py

# Copyright 2019 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http:https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Train a EfficientNets on ImageNet on TPU."""

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

import os
import time
from absl import app
from absl import flags
from absl import logging
import numpy as np
import tensorflow.compat.v1 as tf
from tensorflow.compat.v1 import estimator as tf_estimator
import tensorflow.compat.v2 as tf2  # used for summaries only.

import imagenet_input
import model_builder_factory
import utils
# pylint: disable=g-direct-tensorflow-import
from tensorflow.core.protobuf import rewriter_config_pb2
from tensorflow.python.estimator import estimator
# pylint: enable=g-direct-tensorflow-import

FLAGS = flags.FLAGS

FAKE_DATA_DIR = 'gs:https://cloud-tpu-test-datasets/fake_imagenet'

flags.DEFINE_bool(
    'use_tpu', default=True,
    help=('Use TPU to execute the model for training and evaluation. If'
          ' --use_tpu=false, will use whatever devices are available to'
          ' TensorFlow by default (e.g. CPU and GPU)'))

flags.DEFINE_string('tpu_job_name', None, help=('Name of worker binary.'))

# Cloud TPU Cluster Resolvers
flags.DEFINE_string(
    'tpu', default=None,
    help='The Cloud TPU to use for training. This should be either the name '
    'used when creating the Cloud TPU, or a grpc:https://ip.address.of.tpu:8470 url.')

flags.DEFINE_string(
    'gcp_project', default=None,
    help='Project name for the Cloud TPU-enabled project. If not specified, we '
    'will attempt to automatically detect the GCE project from metadata.')

flags.DEFINE_string(
    'tpu_zone', default=None,
    help='GCE zone where the Cloud TPU is located in. If not specified, we '
    'will attempt to automatically detect the GCE project from metadata.')

# Model specific flags
flags.DEFINE_string(
    'data_dir', default=FAKE_DATA_DIR,
    help=('The directory where the ImageNet input data is stored. Please see'
          ' the README.md for the expected data format.'))

flags.DEFINE_integer(
    'holdout_shards',
    default=None,
    help=('Number of holdout shards for validation. Recommended 20.'))

flags.DEFINE_string('eval_name', default=None, help=('Evaluation name.'))

flags.DEFINE_bool(
    'archive_ckpt', default=True, help=('If true, archive the best ckpt.'))

flags.DEFINE_string(
    'model_dir', default=None,
    help=('The directory where the model and training/evaluation summaries are'
          ' stored.'))

flags.DEFINE_string(
    'model_name',
    default='efficientnet-b0',
    help=('The model name among existing configurations.'))

flags.DEFINE_string(
    'mode', default='train_and_eval',
    help='One of {"train_and_eval", "train", "eval"}.')

flags.DEFINE_string(
    'augment_name', default=None,
    help='`string` that is the name of the augmentation method'
         'to apply to the image. `autoaugment` if AutoAugment is to be used or'
         '`randaugment` if RandAugment is to be used. If the value is `None` no'
         'augmentation method will be applied applied. See autoaugment.py for  '
         'more details.')


flags.DEFINE_integer(
    'randaug_num_layers', default=2,
    help='If RandAug is used, what should the number of layers be.'
         'See autoaugment.py for detailed description.')

flags.DEFINE_integer(
    'randaug_magnitude', default=10,
    help='If RandAug is used, what should the magnitude be. '
         'See autoaugment.py for detailed description.')


flags.DEFINE_integer(
    'train_steps', default=218949,
    help=('The number of steps to use for training. Default is 218949 steps'
          ' which is approximately 350 epochs at batch size 2048. This flag'
          ' should be adjusted according to the --train_batch_size flag.'))

flags.DEFINE_integer(
    'input_image_size', default=None,
    help=('Input image size: it depends on specific model name.'))

flags.DEFINE_integer(
    'train_batch_size', default=2048, help='Batch size for training.')

flags.DEFINE_integer(
    'eval_batch_size', default=64, help='Batch size for evaluation.')

flags.DEFINE_integer(
    'batch_norm_batch_size',
    default=128,
    help='Per-group batch size for distributed batch normalization.')

flags.DEFINE_integer(
    'num_train_images', default=1281167, help='Size of training data set.')

flags.DEFINE_integer(
    'num_eval_images', default=50000, help='Size of evaluation data set.')

flags.DEFINE_integer(
    'steps_per_eval', default=6255,
    help=('Controls how often evaluation is performed. Since evaluation is'
          ' fairly expensive, it is advised to evaluate as infrequently as'
          ' possible (i.e. up to --train_steps, which evaluates the model only'
          ' after finishing the entire training regime).'))

flags.DEFINE_integer(
    'eval_timeout',
    default=None,
    help='Maximum seconds between checkpoints before evaluation terminates.')

flags.DEFINE_bool(
    'skip_host_call', default=False,
    help=('Skip the host_call which is executed every training step. This is'
          ' generally used for generating training summaries (train loss,'
          ' learning rate, etc...). When --skip_host_call=false, there could'
          ' be a performance drop if host_call function is slow and cannot'
          ' keep up with the TPU-side computation.'))

flags.DEFINE_integer(
    'iterations_per_loop', default=1251,
    help=('Number of steps to run on TPU before outfeeding metrics to the CPU.'
          ' If the number of iterations in the loop would exceed the number of'
          ' train steps, the loop will exit before reaching'
          ' --iterations_per_loop. The larger this value is, the higher the'
          ' utilization on the TPU.'))

flags.DEFINE_integer(
    'num_parallel_calls', default=64,
    help=('Number of parallel threads in CPU for the input pipeline'))

flags.DEFINE_string(
    'bigtable_project', None,
    'The Cloud Bigtable project.  If None, --gcp_project will be used.')
flags.DEFINE_string(
    'bigtable_instance', None,
    'The Cloud Bigtable instance to load data from.')
flags.DEFINE_string(
    'bigtable_table', 'imagenet',
    'The Cloud Bigtable table to load data from.')
flags.DEFINE_string(
    'bigtable_train_prefix', 'train_',
    'The prefix identifying training rows.')
flags.DEFINE_string(
    'bigtable_eval_prefix', 'validation_',
    'The prefix identifying evaluation rows.')
flags.DEFINE_string(
    'bigtable_column_family', 'tfexample',
    'The column family storing TFExamples.')
flags.DEFINE_string(
    'bigtable_column_qualifier', 'example',
    'The column name storing TFExamples.')

flags.DEFINE_string(
    'data_format', default='channels_last',
    help=('A flag to override the data format used in the model. The value'
          ' is either channels_first or channels_last. To run the network on'
          ' CPU or TPU, channels_last should be used. For GPU, channels_first'
          ' will improve performance.'))
flags.DEFINE_integer(
    'num_label_classes', default=1000, help='Number of classes, at least 2')

flags.DEFINE_float(
    'batch_norm_momentum',
    default=None,
    help=('Batch normalization layer momentum of moving average to override.'))
flags.DEFINE_float(
    'batch_norm_epsilon',
    default=None,
    help=('Batch normalization layer epsilon to override..'))

flags.DEFINE_bool(
    'transpose_input', default=True,
    help='Use TPU double transpose optimization')

flags.DEFINE_bool(
    'use_bfloat16',
    default=True,
    help=('Whether to use bfloat16 as activation for training.'))

flags.DEFINE_string(
    'export_dir',
    default=None,
    help=('The directory where the exported SavedModel will be stored.'))
flags.DEFINE_bool(
    'export_to_tpu', default=False,
    help=('Whether to export additional metagraph with "serve, tpu" tags'
          ' in addition to "serve" only metagraph.'))

flags.DEFINE_float(
    'base_learning_rate',
    default=0.016,
    help=('Base learning rate when train batch size is 256.'))

flags.DEFINE_float('lr_decay_epoch', default=2.4, help='LR decay epoch.')

flags.DEFINE_float(
    'moving_average_decay', default=0.9999,
    help=('Moving average decay rate.'))

flags.DEFINE_float(
    'weight_decay', default=1e-5,
    help=('Weight decay coefficiant for l2 regularization.'))

flags.DEFINE_float(
    'label_smoothing', default=0.1,
    help=('Label smoothing parameter used in the softmax_cross_entropy'))

flags.DEFINE_float(
    'dropout_rate', default=None,
    help=('Dropout rate for the final output layer.'))

flags.DEFINE_float(
    'survival_prob', default=None,
    help=('Drop connect rate for the network.'))

flags.DEFINE_float(
    'mixup_alpha',
    default=0.0,
    help=('Alpha parameter for mixup regularization, 0.0 to disable.'))

flags.DEFINE_integer('log_step_count_steps', 64, 'The number of steps at '
                     'which the global step information is logged.')

flags.DEFINE_bool(
    'use_cache', default=False, help=('Enable cache for training input.'))

flags.DEFINE_float(
    'depth_coefficient', default=None,
    help=('Depth coefficient for scaling number of layers.'))

flags.DEFINE_float(
    'width_coefficient', default=None,
    help=('Width coefficient for scaling channel size.'))

flags.DEFINE_bool(
    'use_async_checkpointing', default=False, help=('Enable async checkpoint'))

flags.DEFINE_string(
    'optimizer',
    default='rmsprop',
    help='The optimizer to use. Can be either rmsprop, sgd, momentum, or lars.')

flags.DEFINE_string(
    'lr_schedule', default='exponential', help=('learning rate schedule'))

flags.DEFINE_float(
    'lr_decay_factor', default=0.97, help=('Learning rate decay factor.'))

flags.DEFINE_float(
    'lr_warmup_epochs', default=5, help=('warmup epochs for learning rate'))

flags.DEFINE_float(
    'lars_weight_decay',
    default=0.00001,
    help=('Weight decay for LARS optimizer.'))

flags.DEFINE_float(
    'lars_epsilon', default=0.0, help=('Epsilon for LARS optimizer.'))

flags.DEFINE_integer(
    'num_replicas', default=32, help=('Number of TPU replicas.'))


def model_fn(features, labels, mode, params):
  """The model_fn to be used with TPUEstimator.

  Args:
    features: `Tensor` of batched images.
    labels: `Tensor` of one hot labels for the data samples
    mode: one of `tf.estimator.ModeKeys.{TRAIN,EVAL,PREDICT}`
    params: `dict` of parameters passed to the model from the TPUEstimator,
        `params['batch_size']` is always provided and should be used as the
        effective batch size.

  Returns:
    A `TPUEstimatorSpec` for the model
  """
  if isinstance(features, dict):
    features = features['feature']

  # In most cases, the default data format NCHW instead of NHWC should be
  # used for a significant performance boost on GPU. NHWC should be used
  # only if the network needs to be run on CPU since the pooling operations
  # are only supported on NHWC. TPU uses XLA compiler to figure out best layout.
  if FLAGS.data_format == 'channels_first':
    assert not FLAGS.transpose_input    # channels_first only for GPU
    features = tf.transpose(features, [0, 3, 1, 2])
    stats_shape = [3, 1, 1]
  else:
    stats_shape = [1, 1, 3]

  input_image_size = FLAGS.input_image_size
  if not input_image_size:
    input_image_size = model_builder_factory.get_model_input_size(
        FLAGS.model_name)

  if FLAGS.transpose_input and mode != tf_estimator.ModeKeys.PREDICT:
    features = tf.reshape(features,
                          [input_image_size, input_image_size, 3, -1])
    features = tf.transpose(features, [3, 0, 1, 2])  # HWCN to NHWC

  is_training = (mode == tf_estimator.ModeKeys.TRAIN)
  has_moving_average_decay = (FLAGS.moving_average_decay > 0)
  # This is essential, if using a keras-derived model.
  tf.keras.backend.set_learning_phase(is_training)
  logging.info('Using open-source implementation.')
  override_params = {}
  if FLAGS.batch_norm_momentum is not None:
    override_params['batch_norm_momentum'] = FLAGS.batch_norm_momentum
  if FLAGS.batch_norm_epsilon is not None:
    override_params['batch_norm_epsilon'] = FLAGS.batch_norm_epsilon
  if FLAGS.dropout_rate is not None:
    override_params['dropout_rate'] = FLAGS.dropout_rate
  if FLAGS.survival_prob is not None:
    override_params['survival_prob'] = FLAGS.survival_prob
  if FLAGS.data_format:
    override_params['data_format'] = FLAGS.data_format
  if FLAGS.num_label_classes:
    override_params['num_classes'] = FLAGS.num_label_classes
  if FLAGS.depth_coefficient:
    override_params['depth_coefficient'] = FLAGS.depth_coefficient
  if FLAGS.width_coefficient:
    override_params['width_coefficient'] = FLAGS.width_coefficient
  if FLAGS.use_bfloat16:
    override_params['use_bfloat16'] = FLAGS.use_bfloat16

  def normalize_features(features, mean_rgb, stddev_rgb):
    """Normalize the image given the means and stddevs."""
    features -= tf.constant(mean_rgb, shape=stats_shape, dtype=features.dtype)
    features /= tf.constant(stddev_rgb, shape=stats_shape, dtype=features.dtype)
    return features

  def build_model():
    """Build model using the model_name given through the command line."""
    model_builder = model_builder_factory.get_model_builder(FLAGS.model_name)
    normalized_features = normalize_features(features, model_builder.MEAN_RGB,
                                             model_builder.STDDEV_RGB)
    logits, _ = model_builder.build_model(
        normalized_features,
        model_name=FLAGS.model_name,
        training=is_training,
        override_params=override_params,
        model_dir=FLAGS.model_dir)
    return logits

  if params['use_bfloat16']:
    with tf.tpu.bfloat16_scope():
      logits = tf.cast(build_model(), tf.float32)
  else:
    logits = build_model()

  if mode == tf_estimator.ModeKeys.PREDICT:
    predictions = {
        'classes': tf.argmax(logits, axis=1),
        'probabilities': tf.nn.softmax(logits, name='softmax_tensor')
    }
    return tf_estimator.EstimatorSpec(
        mode=mode,
        predictions=predictions,
        export_outputs={
            'classify': tf_estimator.export.PredictOutput(predictions)
        })

  # If necessary, in the model_fn, use params['batch_size'] instead the batch
  # size flags (--train_batch_size or --eval_batch_size).
  batch_size = params['batch_size']   # pylint: disable=unused-variable

  # Calculate loss, which includes softmax cross entropy and L2 regularization.
  cross_entropy = tf.losses.softmax_cross_entropy(
      logits=logits,
      onehot_labels=labels,
      label_smoothing=FLAGS.label_smoothing)

  # Add weight decay to the loss for non-batch-normalization variables.
  loss = cross_entropy + FLAGS.weight_decay * tf.add_n(
      [tf.nn.l2_loss(v) for v in tf.trainable_variables()
       if 'batch_normalization' not in v.name])

  global_step = tf.train.get_global_step()
  if has_moving_average_decay:
    ema = tf.train.ExponentialMovingAverage(
        decay=FLAGS.moving_average_decay, num_updates=global_step)
    ema_vars = utils.get_ema_vars()

  host_call = None
  restore_vars_dict = None
  if is_training:
    # Compute the current epoch and associated learning rate from global_step.
    current_epoch = (
        tf.cast(global_step, tf.float32) / params['steps_per_epoch'])

    scaled_lr = FLAGS.base_learning_rate * (FLAGS.train_batch_size / 256.0)
    logging.info('base_learning_rate = %f', FLAGS.base_learning_rate)
    learning_rate = utils.build_learning_rate(
        scaled_lr,
        global_step,
        params['steps_per_epoch'],
        decay_epochs=FLAGS.lr_decay_epoch,
        warmup_epochs=FLAGS.lr_warmup_epochs,
        decay_factor=FLAGS.lr_decay_factor,
        lr_decay_type=FLAGS.lr_schedule,
        total_steps=FLAGS.train_steps)
    optimizer = utils.build_optimizer(
        learning_rate,
        optimizer_name=FLAGS.optimizer,
        lars_weight_decay=FLAGS.lars_weight_decay,
        lars_epsilon=FLAGS.lars_epsilon)
    if FLAGS.use_tpu:
      # When using TPU, wrap the optimizer with CrossShardOptimizer which
      # handles synchronization details between different TPU cores. To the
      # user, this should look like regular synchronous training.
      optimizer = tf.tpu.CrossShardOptimizer(optimizer)

    # Batch normalization requires UPDATE_OPS to be added as a dependency to
    # the train operation.
    update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
    with tf.control_dependencies(update_ops):
      train_op = optimizer.minimize(loss, global_step)

    if has_moving_average_decay:
      with tf.control_dependencies([train_op]):
        train_op = ema.apply(ema_vars)

    if not FLAGS.skip_host_call:
      def host_call_fn(gs, lr, ce):
        """Training host call. Creates scalar summaries for training metrics.

        This function is executed on the CPU and should not directly reference
        any Tensors in the rest of the `model_fn`. To pass Tensors from the
        model to the `metric_fn`, provide as part of the `host_call`. See
        https://www.tensorflow.org/api_docs/python/tf/estimator/tpu/TPUEstimatorSpec
        for more information.

        Arguments should match the list of `Tensor` objects passed as the second
        element in the tuple passed to `host_call`.

        Args:
          gs: `Tensor with shape `[batch]` for the global_step
          lr: `Tensor` with shape `[batch]` for the learning_rate.
          ce: `Tensor` with shape `[batch]` for the current_epoch.

        Returns:
          List of summary ops to run on the CPU host.
        """
        gs = gs[0]
        # Host call fns are executed FLAGS.iterations_per_loop times after one
        # TPU loop is finished, setting max_queue value to the same as number of
        # iterations will make the summary writer only flush the data to storage
        # once per loop.
        with tf2.summary.create_file_writer(
            FLAGS.model_dir, max_queue=FLAGS.iterations_per_loop).as_default():
          with tf2.summary.record_if(True):
            tf2.summary.scalar('learning_rate', lr[0], step=gs)
            tf2.summary.scalar('current_epoch', ce[0], step=gs)

            return tf.summary.all_v2_summary_ops()

      # To log the loss, current learning rate, and epoch for Tensorboard, the
      # summary op needs to be run on the host CPU via host_call. host_call
      # expects [batch_size, ...] Tensors, thus reshape to introduce a batch
      # dimension. These Tensors are implicitly concatenated to
      # [params['batch_size']].
      gs_t = tf.reshape(global_step, [1])
      lr_t = tf.reshape(learning_rate, [1])
      ce_t = tf.reshape(current_epoch, [1])

      host_call = (host_call_fn, [gs_t, lr_t, ce_t])

  else:
    train_op = None
    if has_moving_average_decay:
      # Load moving average variables for eval.
      restore_vars_dict = ema.variables_to_restore(ema_vars)

  eval_metrics = None
  if mode == tf_estimator.ModeKeys.EVAL:
    def metric_fn(labels, logits):
      """Evaluation metric function. Evaluates accuracy.

      This function is executed on the CPU and should not directly reference
      any Tensors in the rest of the `model_fn`. To pass Tensors from the model
      to the `metric_fn`, provide as part of the `eval_metrics`. See
      https://www.tensorflow.org/api_docs/python/tf/estimator/tpu/TPUEstimatorSpec
      for more information.

      Arguments should match the list of `Tensor` objects passed as the second
      element in the tuple passed to `eval_metrics`.

      Args:
        labels: `Tensor` with shape `[batch, num_classes]`.
        logits: `Tensor` with shape `[batch, num_classes]`.

      Returns:
        A dict of the metrics to return from evaluation.
      """
      labels = tf.argmax(labels, axis=1)
      predictions = tf.argmax(logits, axis=1)
      top_1_accuracy = tf.metrics.accuracy(labels, predictions)
      in_top_5 = tf.cast(tf.nn.in_top_k(logits, labels, 5), tf.float32)
      top_5_accuracy = tf.metrics.mean(in_top_5)

      return {
          'top_1_accuracy': top_1_accuracy,
          'top_5_accuracy': top_5_accuracy,
      }

    eval_metrics = (metric_fn, [labels, logits])

  num_params = np.sum([np.prod(v.shape) for v in tf.trainable_variables()])
  logging.info('number of trainable parameters: %d', num_params)

  def _scaffold_fn():
    saver = tf.train.Saver(restore_vars_dict)
    return tf.train.Scaffold(saver=saver)

  if has_moving_average_decay and not is_training:
    # Only apply scaffold for eval jobs.
    scaffold_fn = _scaffold_fn
  else:
    scaffold_fn = None

  return tf_estimator.tpu.TPUEstimatorSpec(
      mode=mode,
      loss=loss,
      train_op=train_op,
      host_call=host_call,
      eval_metrics=eval_metrics,
      scaffold_fn=scaffold_fn)


def _verify_non_empty_string(value, field_name):
  """Ensures that a given proposed field value is a non-empty string.

  Args:
    value:  proposed value for the field.
    field_name:  string name of the field, e.g. `project`.

  Returns:
    The given value, provided that it passed the checks.

  Raises:
    ValueError:  the value is not a string, or is a blank string.
  """
  if not isinstance(value, str):
    raise ValueError(
        'Bigtable parameter "%s" must be a string.' % field_name)
  if not value:
    raise ValueError(
        'Bigtable parameter "%s" must be non-empty.' % field_name)
  return value


def _select_tables_from_flags():
  """Construct training and evaluation Bigtable selections from flags.

  Returns:
    [training_selection, evaluation_selection]
  """
  project = _verify_non_empty_string(
      FLAGS.bigtable_project or FLAGS.gcp_project,
      'project')
  instance = _verify_non_empty_string(FLAGS.bigtable_instance, 'instance')
  table = _verify_non_empty_string(FLAGS.bigtable_table, 'table')
  train_prefix = _verify_non_empty_string(FLAGS.bigtable_train_prefix,
                                          'train_prefix')
  eval_prefix = _verify_non_empty_string(FLAGS.bigtable_eval_prefix,
                                         'eval_prefix')
  column_family = _verify_non_empty_string(FLAGS.bigtable_column_family,
                                           'column_family')
  column_qualifier = _verify_non_empty_string(FLAGS.bigtable_column_qualifier,
                                              'column_qualifier')
  return [
      imagenet_input.BigtableSelection(  # pylint: disable=g-complex-comprehension
          project=project,
          instance=instance,
          table=table,
          prefix=p,
          column_family=column_family,
          column_qualifier=column_qualifier)
      for p in (train_prefix, eval_prefix)
  ]


def export(est, export_dir, input_image_size=None):
  """Export graph to SavedModel and TensorFlow Lite.

  Args:
    est: estimator instance.
    export_dir: string, exporting directory.
    input_image_size: int, input image size.

  Raises:
    ValueError: the export directory path is not specified.
  """
  if not export_dir:
    raise ValueError('The export directory path is not specified.')

  if not input_image_size:
    input_image_size = FLAGS.input_image_size
  is_cond_conv = FLAGS.model_name.startswith('efficientnet-condconv')
  batch_size = 1 if is_cond_conv else None  # Use fixed batch size for condconv.

  logging.info('Starting to export model.')
  if (FLAGS.model_name.startswith('efficientnet-lite') or
      FLAGS.model_name.startswith('efficientnet-edgetpu')):
    # lite or edgetpu use binlinear for easier post-quantization.
    resize_method = tf.image.ResizeMethod.BILINEAR
  else:
    resize_method = None
  image_serving_input_fn = imagenet_input.build_image_serving_input_fn(
      input_image_size, batch_size=batch_size, resize_method=resize_method)
  est.export_saved_model(
      export_dir_base=export_dir,
      serving_input_receiver_fn=image_serving_input_fn)


def main(unused_argv):

  input_image_size = FLAGS.input_image_size
  if not input_image_size:
    input_image_size = model_builder_factory.get_model_input_size(
        FLAGS.model_name)

  if FLAGS.holdout_shards:
    holdout_images = int(FLAGS.num_train_images * FLAGS.holdout_shards / 1024.0)
    FLAGS.num_train_images -= holdout_images
    if FLAGS.eval_name and 'test' in FLAGS.eval_name:
      FLAGS.holdout_shards = None  # do not use holdout if eval test set.
    else:
      FLAGS.num_eval_images = holdout_images

  # For imagenet dataset, include background label if number of output classes
  # is 1001
  include_background_label = (FLAGS.num_label_classes == 1001)

  if FLAGS.tpu or FLAGS.use_tpu:
    tpu_cluster_resolver = tf.distribute.cluster_resolver.TPUClusterResolver(
        FLAGS.tpu,
        zone=FLAGS.tpu_zone,
        project=FLAGS.gcp_project)
  else:
    tpu_cluster_resolver = None

  if FLAGS.use_async_checkpointing:
    save_checkpoints_steps = None
  else:
    save_checkpoints_steps = max(100, FLAGS.iterations_per_loop)
  config = tf_estimator.tpu.RunConfig(
      cluster=tpu_cluster_resolver,
      model_dir=FLAGS.model_dir,
      save_checkpoints_steps=save_checkpoints_steps,
      log_step_count_steps=FLAGS.log_step_count_steps,
      session_config=tf.ConfigProto(
          graph_options=tf.GraphOptions(
              rewrite_options=rewriter_config_pb2.RewriterConfig(
                  disable_meta_optimizer=True))),
      tpu_config=tf_estimator.tpu.TPUConfig(
          iterations_per_loop=FLAGS.iterations_per_loop,
          tpu_job_name=FLAGS.tpu_job_name,
          per_host_input_for_training=tf_estimator.tpu.InputPipelineConfig
          .PER_HOST_V2))  # pylint: disable=line-too-long
  # Initializes model parameters.
  params = dict(
      steps_per_epoch=FLAGS.num_train_images / FLAGS.train_batch_size,
      use_bfloat16=FLAGS.use_bfloat16)
  est = tf_estimator.tpu.TPUEstimator(
      use_tpu=FLAGS.use_tpu,
      model_fn=model_fn,
      config=config,
      train_batch_size=FLAGS.train_batch_size,
      eval_batch_size=FLAGS.eval_batch_size,
      export_to_tpu=FLAGS.export_to_tpu,
      params=params)

  if (FLAGS.model_name.startswith('efficientnet-lite') or
      FLAGS.model_name.startswith('efficientnet-edgetpu')):
    # lite or edgetpu use binlinear for easier post-quantization.
    resize_method = tf.image.ResizeMethod.BILINEAR
  else:
    resize_method = None
  # Input pipelines are slightly different (with regards to shuffling and
  # preprocessing) between training and evaluation.
  def build_imagenet_input(is_training):
    """Generate ImageNetInput for training and eval."""
    if FLAGS.bigtable_instance:
      logging.info('Using Bigtable dataset, table %s', FLAGS.bigtable_table)
      select_train, select_eval = _select_tables_from_flags()
      return imagenet_input.ImageNetBigtableInput(
          is_training=is_training,
          use_bfloat16=FLAGS.use_bfloat16,
          transpose_input=FLAGS.transpose_input,
          selection=select_train if is_training else select_eval,
          num_label_classes=FLAGS.num_label_classes,
          include_background_label=include_background_label,
          augment_name=FLAGS.augment_name,
          mixup_alpha=FLAGS.mixup_alpha,
          randaug_num_layers=FLAGS.randaug_num_layers,
          randaug_magnitude=FLAGS.randaug_magnitude,
          resize_method=resize_method)
    else:
      if FLAGS.data_dir == FAKE_DATA_DIR:
        logging.info('Using fake dataset.')
      else:
        logging.info('Using dataset: %s', FLAGS.data_dir)

      return imagenet_input.ImageNetInput(
          is_training=is_training,
          data_dir=FLAGS.data_dir,
          transpose_input=FLAGS.transpose_input,
          cache=FLAGS.use_cache and is_training,
          image_size=input_image_size,
          num_parallel_calls=FLAGS.num_parallel_calls,
          use_bfloat16=FLAGS.use_bfloat16,
          num_label_classes=FLAGS.num_label_classes,
          include_background_label=include_background_label,
          augment_name=FLAGS.augment_name,
          mixup_alpha=FLAGS.mixup_alpha,
          randaug_num_layers=FLAGS.randaug_num_layers,
          randaug_magnitude=FLAGS.randaug_magnitude,
          resize_method=resize_method,
          holdout_shards=FLAGS.holdout_shards)

  imagenet_train = build_imagenet_input(is_training=True)
  imagenet_eval = build_imagenet_input(is_training=False)

  if FLAGS.mode == 'eval':
    eval_steps = FLAGS.num_eval_images // FLAGS.eval_batch_size
    # Run evaluation when there's a new checkpoint
    for ckpt in tf.train.checkpoints_iterator(
        FLAGS.model_dir, timeout=FLAGS.eval_timeout):
      logging.info('Starting to evaluate.')
      try:
        start_timestamp = time.time()  # This time will include compilation time
        eval_results = est.evaluate(
            input_fn=imagenet_eval.input_fn,
            steps=eval_steps,
            checkpoint_path=ckpt,
            name=FLAGS.eval_name)
        elapsed_time = int(time.time() - start_timestamp)
        logging.info('Eval results: %s. Elapsed seconds: %d',
                     eval_results, elapsed_time)
        if FLAGS.archive_ckpt:
          utils.archive_ckpt(eval_results, eval_results['top_1_accuracy'], ckpt)

        # Terminate eval job when final checkpoint is reached
        try:
          current_step = int(os.path.basename(ckpt).split('-')[1])
        except IndexError:
          logging.info('%s has no global step info: stop!', ckpt)
          break

        if current_step >= FLAGS.train_steps:
          logging.info(
              'Evaluation finished after training step %d', current_step)
          break

      except tf.errors.NotFoundError:
        # Since the coordinator is on a different job than the TPU worker,
        # sometimes the TPU worker does not finish initializing until long after
        # the CPU job tells it to start evaluating. In this case, the checkpoint
        # file could have been deleted already.
        logging.info(
            'Checkpoint %s no longer exists, skipping checkpoint', ckpt)
  else:   # FLAGS.mode == 'train' or FLAGS.mode == 'train_and_eval'
    current_step = estimator._load_global_step_from_checkpoint_dir(FLAGS.model_dir)  # pylint: disable=protected-access,line-too-long

    logging.info(
        'Training for %d steps (%.2f epochs in total). Current'
        ' step %d.', FLAGS.train_steps,
        FLAGS.train_steps / params['steps_per_epoch'], current_step)

    start_timestamp = time.time()  # This time will include compilation time

    if FLAGS.mode == 'train':
      hooks = []
      if FLAGS.use_async_checkpointing:
        try:
          from tensorflow.contrib.tpu.python.tpu import async_checkpoint  # pylint: disable=g-import-not-at-top
        except ImportError as e:
          logging.exception(
              'Async checkpointing is not supported in TensorFlow 2.x')
          raise e

        hooks.append(
            async_checkpoint.AsyncCheckpointSaverHook(
                checkpoint_dir=FLAGS.model_dir,
                save_steps=max(100, FLAGS.iterations_per_loop)))
      est.train(
          input_fn=imagenet_train.input_fn,
          max_steps=FLAGS.train_steps,
          hooks=hooks)

    else:
      assert FLAGS.mode == 'train_and_eval'
      while current_step < FLAGS.train_steps:
        # Train for up to steps_per_eval number of steps.
        # At the end of training, a checkpoint will be written to --model_dir.
        next_checkpoint = min(current_step + FLAGS.steps_per_eval,
                              FLAGS.train_steps)
        est.train(input_fn=imagenet_train.input_fn, max_steps=next_checkpoint)
        current_step = next_checkpoint

        logging.info('Finished training up to step %d. Elapsed seconds %d.',
                     next_checkpoint, int(time.time() - start_timestamp))

        # Evaluate the model on the most recent model in --model_dir.
        # Since evaluation happens in batches of --eval_batch_size, some images
        # may be excluded modulo the batch size. As long as the batch size is
        # consistent, the evaluated images are also consistent.
        logging.info('Starting to evaluate.')
        eval_results = est.evaluate(
            input_fn=imagenet_eval.input_fn,
            steps=FLAGS.num_eval_images // FLAGS.eval_batch_size,
            name=FLAGS.eval_name)
        logging.info('Eval results at step %d: %s',
                     next_checkpoint, eval_results)
        ckpt = tf.train.latest_checkpoint(FLAGS.model_dir)
        if FLAGS.archive_ckpt:
          utils.archive_ckpt(eval_results, eval_results['top_1_accuracy'], ckpt)

      elapsed_time = int(time.time() - start_timestamp)
      logging.info('Finished training up to step %d. Elapsed seconds %d.',
                   FLAGS.train_steps, elapsed_time)
  if FLAGS.export_dir:
    export(est, FLAGS.export_dir, input_image_size)


if __name__ == '__main__':
  logging.set_verbosity(logging.INFO)
  app.run(main)