Python resnet_model.ResNet方法代码示例

# 需要导入模块: import resnet_model [as 别名]
# 或者: from resnet_model import ResNet [as 别名]
def get_model(hps, dataset, train_data_path, mode='train'):
  images, labels = cifar_input.build_input(
    dataset, train_data_path, hps.batch_size, mode)
  model = resnet_model.ResNet(hps, images, labels, mode)
  model.build_graph()
  return model 

# 需要导入模块: import resnet_model [as 别名]
# 或者: from resnet_model import ResNet [as 别名]
def evaluate(hps):
  """Eval loop."""
  images, labels = cifar_input.build_input(
      FLAGS.dataset, FLAGS.eval_data_path, hps.batch_size, FLAGS.mode)
  model = resnet_model.ResNet(hps, images, labels, FLAGS.mode)
  model.build_graph()
  saver = tf.train.Saver()
  summary_writer = tf.summary.FileWriter(FLAGS.eval_dir)

  sess = tf.Session(config=tf.ConfigProto(allow_soft_placement=True))
  tf.train.start_queue_runners(sess)

  best_precision = 0.0
  while True:
    try:
      ckpt_state = tf.train.get_checkpoint_state(FLAGS.log_root)
    except tf.errors.OutOfRangeError as e:
      tf.logging.error('Cannot restore checkpoint: %s', e)
      continue
    if not (ckpt_state and ckpt_state.model_checkpoint_path):
      tf.logging.info('No model to eval yet at %s', FLAGS.log_root)
      continue
    tf.logging.info('Loading checkpoint %s', ckpt_state.model_checkpoint_path)
    saver.restore(sess, ckpt_state.model_checkpoint_path)

    total_prediction, correct_prediction = 0, 0
    for _ in six.moves.range(FLAGS.eval_batch_count):
      (summaries, loss, predictions, truth, train_step) = sess.run(
          [model.summaries, model.cost, model.predictions,
           model.labels, model.global_step])

      truth = np.argmax(truth, axis=1)
      predictions = np.argmax(predictions, axis=1)
      correct_prediction += np.sum(truth == predictions)
      total_prediction += predictions.shape[0]

    precision = 1.0 * correct_prediction / total_prediction
    best_precision = max(precision, best_precision)

    precision_summ = tf.Summary()
    precision_summ.value.add(
        tag='Precision', simple_value=precision)
    summary_writer.add_summary(precision_summ, train_step)
    best_precision_summ = tf.Summary()
    best_precision_summ.value.add(
        tag='Best Precision', simple_value=best_precision)
    summary_writer.add_summary(best_precision_summ, train_step)
    summary_writer.add_summary(summaries, train_step)
    tf.logging.info('loss: %.3f, precision: %.3f, best precision: %.3f' %
                    (loss, precision, best_precision))
    summary_writer.flush()

    if FLAGS.eval_once:
      break

    time.sleep(60) 

# 需要导入模块: import resnet_model [as 别名]
# 或者: from resnet_model import ResNet [as 别名]
def train(hps):
  """Training loop."""
  images, labels = synthetic_data(hps.batch_size)
  model = resnet_model.ResNet(hps, images, labels, FLAGS.mode)
  model.build_graph()
  summary_writer = tf.train.SummaryWriter(FLAGS.train_dir)

  sv = tf.train.Supervisor(logdir=FLAGS.log_root,
                           is_chief=True,
                           summary_op=None,
                           save_summaries_secs=60,
                           save_model_secs=300,
                           global_step=model.global_step)
  sess = sv.prepare_or_wait_for_session(
      config=tf.ConfigProto(allow_soft_placement=True))

  step = 0
  lrn_rate = 0.1

  while not sv.should_stop():
    (_, summaries, loss, predictions, truth, train_step) = sess.run(
        [model.train_op, model.summaries, model.cost, model.predictions,
         model.labels, model.global_step],
        feed_dict={model.lrn_rate: lrn_rate})

    if train_step < 40000:
      lrn_rate = 0.1
    elif train_step < 60000:
      lrn_rate = 0.01
    elif train_step < 80000:
      lrn_rate = 0.001
    else:
      lrn_rate = 0.0001

    truth = np.argmax(truth, axis=1)
    predictions = np.argmax(predictions, axis=1)
    precision = np.mean(truth == predictions)

    step += 1
    if step % 100 == 0:
      precision_summ = tf.Summary()
      precision_summ.value.add(
          tag='Precision', simple_value=precision)
      summary_writer.add_summary(precision_summ, train_step)
      summary_writer.add_summary(summaries, train_step)
      tf.logging.info('loss: %.3f, precision: %.3f\n' % (loss, precision))
      summary_writer.flush()

  sv.Stop() 

# 需要导入模块: import resnet_model [as 别名]
# 或者: from resnet_model import ResNet [as 别名]
def evaluate(hps):
  """Eval loop."""
  images, labels = cifar_input.build_input(
      FLAGS.dataset, FLAGS.eval_data_path, hps.batch_size, FLAGS.mode)
  model = resnet_model.ResNet(hps, images, labels, FLAGS.mode)
  model.build_graph()
  saver = tf.train.Saver()
  summary_writer = tf.train.SummaryWriter(FLAGS.eval_dir)

  sess = tf.Session(config=tf.ConfigProto(allow_soft_placement=True))
  tf.train.start_queue_runners(sess)

  best_precision = 0.0
  while True:
    time.sleep(60)
    try:
      ckpt_state = tf.train.get_checkpoint_state(FLAGS.log_root)
    except tf.errors.OutOfRangeError as e:
      tf.logging.error('Cannot restore checkpoint: %s', e)
      continue
    if not (ckpt_state and ckpt_state.model_checkpoint_path):
      tf.logging.info('No model to eval yet at %s', FLAGS.log_root)
      continue
    tf.logging.info('Loading checkpoint %s', ckpt_state.model_checkpoint_path)
    saver.restore(sess, ckpt_state.model_checkpoint_path)

    total_prediction, correct_prediction = 0, 0
    for _ in xrange(FLAGS.eval_batch_count):
      (summaries, loss, predictions, truth, train_step) = sess.run(
          [model.summaries, model.cost, model.predictions,
           model.labels, model.global_step])

      truth = np.argmax(truth, axis=1)
      predictions = np.argmax(predictions, axis=1)
      correct_prediction += np.sum(truth == predictions)
      total_prediction += predictions.shape[0]

    precision = 1.0 * correct_prediction / total_prediction
    best_precision = max(precision, best_precision)

    precision_summ = tf.Summary()
    precision_summ.value.add(
        tag='Precision', simple_value=precision)
    summary_writer.add_summary(precision_summ, train_step)
    best_precision_summ = tf.Summary()
    best_precision_summ.value.add(
        tag='Best Precision', simple_value=best_precision)
    summary_writer.add_summary(best_precision_summ, train_step)
    summary_writer.add_summary(summaries, train_step)
    tf.logging.info('loss: %.3f, precision: %.3f, best precision: %.3f\n' %
                    (loss, precision, best_precision))
    summary_writer.flush()

    if FLAGS.eval_once:
      break 

# 需要导入模块: import resnet_model [as 别名]
# 或者: from resnet_model import ResNet [as 别名]
def train(hps):
  """Training loop."""
  images, labels = cifar_input.build_input(
      FLAGS.dataset, FLAGS.train_data_path, hps.batch_size, FLAGS.mode)
  model = resnet_model.ResNet(hps, images, labels, FLAGS.mode)
  model.build_graph()
  summary_writer = tf.train.SummaryWriter(FLAGS.train_dir)

  sv = tf.train.Supervisor(logdir=FLAGS.log_root,
                           is_chief=True,
                           summary_op=None,
                           save_summaries_secs=60,
                           save_model_secs=300,
                           global_step=model.global_step)
  sess = sv.prepare_or_wait_for_session(
      config=tf.ConfigProto(allow_soft_placement=True))

  step = 0
  lrn_rate = 0.1

  while not sv.should_stop():
    (_, summaries, loss, predictions, truth, train_step) = sess.run(
        [model.train_op, model.summaries, model.cost, model.predictions,
         model.labels, model.global_step],
        feed_dict={model.lrn_rate: lrn_rate})

    if train_step < 40000:
      lrn_rate = 0.1
    elif train_step < 60000:
      lrn_rate = 0.01
    elif train_step < 80000:
      lrn_rate = 0.001
    else:
      lrn_rate = 0.0001

    truth = np.argmax(truth, axis=1)
    predictions = np.argmax(predictions, axis=1)
    precision = np.mean(truth == predictions)

    step += 1
    if step % 100 == 0:
      precision_summ = tf.Summary()
      precision_summ.value.add(
          tag='Precision', simple_value=precision)
      summary_writer.add_summary(precision_summ, train_step)
      summary_writer.add_summary(summaries, train_step)
      tf.logging.info('loss: %.3f, precision: %.3f\n' % (loss, precision))
      summary_writer.flush()

  sv.Stop() 

# 需要导入模块: import resnet_model [as 别名]
# 或者: from resnet_model import ResNet [as 别名]
def eval_resnet():
  """Evaluates the resnet model."""
  if not os.path.exists(FLAGS.eval_dir):
    os.makedirs(FLAGS.eval_dir)
  g = tf.Graph()
  with g.as_default():
    # pylint: disable=line-too-long
    images, one_hot_labels, num_samples, num_of_classes = cifar_data_provider.provide_resnet_data(
        FLAGS.dataset_name,
        FLAGS.split_name,
        FLAGS.batch_size,
        dataset_dir=FLAGS.data_dir,
        num_epochs=None)

    hps = resnet_model.HParams(
        batch_size=FLAGS.batch_size,
        num_classes=num_of_classes,
        min_lrn_rate=0.0001,
        lrn_rate=0,
        num_residual_units=9,
        use_bottleneck=False,
        weight_decay_rate=0.0002,
        relu_leakiness=0.1,
        optimizer='mom')

    # Define the model:
    images.set_shape([FLAGS.batch_size, 32, 32, 3])
    resnet = resnet_model.ResNet(hps, images, one_hot_labels, mode='test')

    logits = resnet.build_model()

    total_loss = tf.nn.softmax_cross_entropy_with_logits(
        labels=one_hot_labels, logits=logits)
    total_loss = tf.reduce_mean(total_loss, name='xent')

    slim.summaries.add_scalar_summary(
        total_loss, 'total_loss', print_summary=True)

    # Define the metrics:
    predictions = tf.argmax(logits, 1)
    labels = tf.argmax(one_hot_labels, 1)

    names_to_values, names_to_updates = slim.metrics.aggregate_metric_map({
        'accuracy': tf.metrics.accuracy(predictions, labels),
    })

    for name, value in names_to_values.iteritems():
      slim.summaries.add_scalar_summary(
          value, name, prefix='eval', print_summary=True)

    # This ensures that we make a single pass over all of the data.
    num_batches = math.ceil(num_samples / float(FLAGS.batch_size))

    slim.evaluation.evaluation_loop(
        master=FLAGS.master,
        checkpoint_dir=FLAGS.checkpoint_dir,
        logdir=FLAGS.eval_dir,
        num_evals=num_batches,
        eval_op=names_to_updates.values(),
        eval_interval_secs=FLAGS.eval_interval_secs)