Question

用于评估经过预训练的模型（deepLab），我使用了Frozen_graph建立了模型。但是我无法加载权重并正确初始化模型。我的火车目录中有一些文件，例如（frozen.Pb-* .meta-ckpt），但我不能使用它们

这是我的代码

class DeepLabModel2(object):

  INPUT_TENSOR_NAME = 'ImageTensor:0'
  OUTPUT_TENSOR_NAME = 'semantic:0' #'ArgMax:0'#'SemanticPredictions:0'
  INPUT_SIZE = 513
  num_classes = 21;
  FROZEN_GRAPH_NAME = 'frozen_inference_graph'

  def __init__(self, tarball_path):
      self.graph = tf.Graph();
      with self.graph.as_default():
          model_options = common.ModelOptions(
              outputs_to_num_classes={common.OUTPUT_TYPE: self.num_classes},
              crop_size=FLAGS.eval_crop_size,
              atrous_rates=FLAGS.atrous_rates,
              output_stride=FLAGS.output_stride)

          tf.logging.info('Performing single-scale test.')
          samples = tf.placeholder(tf.float32, shape=(None, self.INPUT_SIZE, self.INPUT_SIZE, 3), name='ImageTensor')
          self.predictions = model.predict_labels(samples, model_options,
                                             image_pyramid=FLAGS.image_pyramid)
          #predictions2 = predictions[common.OUTPUT_TYPE]
          #predictions = tf.reshape(predictions, shape=[-1])
          self.sess = tf.Session(graph=self.graph)
          tf.global_variables_initializer().run(session=self.sess);

我使用以下代码恢复体重

saver = tf.train.Saver(variables_to_restore)

with tf.Session() as self.sess:
    saver.restore(self.sess, model_path)

并出现错误：ValueError：没有要保存的变量

如何在张量流中加载图可变权重

0 个答案: