tensorflow saver.restore

tf_saver.py

def train():
  with tf.Graph().as_default():
    # 初期化オペレーション
    init_op = tf.initialize_all_variables()

    # Session
    sess = tf.Session(config=tf.ConfigProto(log_device_placement=LOG_DEVICE_PLACEMENT))

    sess.run(init_op)
    all_variables = tf.all_variables()
    
    # filterd variables by name, parameters list that I need.
    scale1and2_params = []
    for variable in all_variables:
      # ...
      scale1and2_params.append(variable)
    
    saver_scale1and2 = tf.train.Saver(scale1and2_params)
    
    # load checkpoint from CloudStorage
    # FLAGS.train_s12_dir is "gs://<projectid>/cnn_depth_labels/<pretrained jobid>/train12"
    scale1and2_ckpt = tf.train.get_checkpoint_state(FLAGS.train_s12_dir)
    if scale1and2_ckpt and scale1and2_ckpt.model_checkpoint_path:
      # scale1and2_ckpt.model_checkpoint_path is "gs://<projectid>-ml/cnn_depth_labels/<pretrained jobid>/train12/model.ckpt-90"
      print("Pretrained scale1and2 Model Loading. %s" % (scale1and2_ckpt.model_checkpoint_path))
      saver_scale1and2.restore(sess, scale1and2_ckpt.model_checkpoint_path)
      # ->>> exception