python - 值错误: No variables to save when saving Tensorflow checkpoint

Question

我一直在尝试使用 freeze_graph函数来获取模型+权重/偏差，但在这个过程中，我发现我的初始网络似乎没有任何变量，尽管能够正确分类图像。我的代码如下:

#!/usr/bin/python
import tensorflow as tf
import numpy as np


def outputCheckpoint(sess):
    with sess.as_default():
        print("Saving to checkpoint")
        saver = tf.train.Saver()
        # Fails on this line: 'ValueError: No variables to save'
        saver.save(sess, '/path/to/tensorflow/graph_saver/save_checkpoint')

def main():
    with tf.device('/cpu:0'):
        with open("tensorflow_inception_graph.pb", mode='rb') as f:
            fileContent = f.read()

        graph_def = tf.GraphDef()
        graph_def.ParseFromString(fileContent)

        images = tf.placeholder("float", [ None, 299, 299, 3])

        tf.import_graph_def(graph_def, input_map={ "Mul": images })
        print "graph loaded from disk"

        graph = tf.get_default_graph()

        with tf.Session() as sess:
            init = tf.global_variables_initializer()
            sess.run(init)
            outputCheckpoint(sess)

main()

错误是:

graph loaded from disk
Saving to checkpoint
Traceback (most recent call last):
  File "./inception_benchmark.py", line 28, in <module>
    main()
  File "./inception_benchmark.py", line 24, in main
    saver = tf.train.Saver()
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/training/saver.py", line 1067, in __init__
    self.build()
  File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/training/saver.py", line 1088, in build
    raise ValueError("No variables to save")
ValueError: No variables to save

如果您尚未下载 inception 网络，这将为您提供 .pb 文件:

$ wget https://storage.googleapis.com/download.tensorflow.org/models/inception_dec_2015.zip -O tensorflow/examples/label_image/data/inception_dec_2015.zip

另外，如果有人好奇的话，这是我的完整代码的要点: gist

有人知道这是怎么回事吗？

谢谢!

Answer 1

freeze_graph 工具的作用是将图表中的所有 Variable 节点转换为 tf.constant 节点 - 这允许您保存在单个 protobuf 文件中对图运行推理所需的所有信息(而不是分别保存 GraphDef 和 Variable 检查点数据)。

您在这里遇到的是成功卡住图(tensorflow_inception_graph.pb)的结果:没有要保存的变量，因为它们已全部转换为常量!