ValueError:节点的输入0与预期的float_ref不兼容.**

Question

我在尝试导入优化的冻结图时遇到异常.

# read pb into graph_def
with tf.gfile.GFile(pb_file, "rb") as f:
    graph_def = tf.GraphDef()
    graph_def.ParseFromString(f.read())

# import graph_def
with tf.Graph().as_default() as graph:
    tf.import_graph_def(graph_def)

在此行中获取异常:

tf.import_graph_def(graph_def)

回溯(最近通话最后一个):文件
"/home/automator/PycharmProjects/tensorflow/venv/lib/python3.5/site-
包/ tensorflow /蟒蛇/框架/ importer.py",线路489,在
import_graph_def graph._c_graph ,序列化,选项)#pylint:disable = protected-access
tensorflow.python.framework.errors_impl.InvalidArgumentError:节点
import/final_retrain_ops/Wx_plus_b/weights_quant/AssignMinLast的输入0
从
import/final_retrain_ops/Wx_plus_b/weights_quant/min 传递给float : 0
与预期的float_ref 不兼容.在处理上述异常,另一个异常:回溯(最近通话最后一个):
文件"/snap/pycharm-community/64/helpers/pydev/pydevd.py",行1664,在main()文件"/ SNAP /pycharm-community/64/helpers/pydev/pydevd.py "线1658,在
主全局= debugger.run(设定[ '文件'],无,无,is_module)文件"/ SNAP/pycharm社区/ 64 /helpers/pydev/pydevd.py",第1068行,运行pydev_imports.execfile(文件,全局,本地)#执行脚本文件
"/snap/pycharm-community/64/helpers/pydev/_pydev_imps/_pydev_execfile.py"第18行,在EXEC的execfile(编译(内容+ "\n"个,文件, 'EXEC'),水珠,LOC)文件"/home/automator/PycharmProjects/tensorflow/tfliteme.py",线路389,
在printTensors(" /home/automator/Desktop/cervix/optimized_model.pb")
文件"/home/automator/PycharmProjects/tensorflow/tfliteme.py",第
374 行,printTensors中的tf.import_graph_def(graph_def)文件"/ home/automator/PycharmProjects/tensorflow/venv/lib/python3.5/site-
packages/tensorflow/python/util/deprecation .py",第432行,在
new_func中返回func(*args,**kwargs)文件"/home/automator/PycharmProjects/tensorflow/venv/lib/python3.5/site-
packages/tensorflow/python/framework/importer. PY",线路493,在
import_graph_def提高ValueError异常(STR(e))的ValueError异常:节点进口/ final_retrain_ops/Wx_plus_b/weights_quant/AssignMinLast的输入0是
通过从浮
进口/ final_retrain_ops/Wx_plus_b/weights_quant /分钟:0不相容
与

期望的float_ref.

Answer 1

确保您的pb_file格式正确（类似这样），并尝试在“名称”参数中使用一些值来import_graph_def()尝试覆盖“导入”默认值，如下所示：

# read pb into graph_def
with tf.gfile.GFile(pb_file, "rb") as f:
    graph_def = tf.GraphDef()
    graph_def.ParseFromString(f.read())

# import graph_def
with tf.Graph().as_default() as graph:
    tf.import_graph_def(graph_def, name='')