我正在使用TensorFlow在python中使用经过训练的模型。并且我需要将tensorflow格式更改为ONNX,但是当我使用此命令时:
(python -m tf2onnx.convert --graphdef。\ saved_model.pb-输出冻结的.onnx --fold_const --opset 8-输入input_node:0-输出output_node:0)
然后在cmd中,出现此错误:
tensorflow.python.framework.errors_impl.FailedPreconditionError:尝试使用未初始化的值Variable_1 [[{{node _retval_Variable_1_0_3}}]]
我知道此错误意味着我正在使用一个未初始化的变量,但实际上我无法理解它是什么变量,即使我从变量中得到了一个列表,如下所示: (此外,训练模型时没有出现任何错误。)
'''
W_conv1 RefVariable <tf.Variable 'Variable:0'<...>1, 32)
dtype=float32_ref>
W_conv2 RefVariable <tf.Variable 'Variable_2:<...>2, 64)
dtype=float32_ref>
W_conv3 RefVariable <tf.Variable 'Variable_4:<...>, 128)
dtype=float32_ref>
W_conv4 RefVariable <tf.Variable 'Variable_6:<...>, 256)
dtype=float32_ref>
W_fc1 RefVariable <tf.Variable 'Variable_8:<...> 9216)
dtype=float32_ref>
W_fc2 RefVariable <tf.Variable 'Variable_10<...>16, 1)
dtype=float32_ref>
absolute_import _Feature _Feature((2, 5, 0, 'alpha<...>0, 0,
'alpha', 0), 16384)
b_conv1 RefVariable <tf.Variable 'Variable_1:<...>=(32,)
dtype=float32_ref>
b_conv2 RefVariable <tf.Variable 'Variable_3:<...>=(64,)
dtype=float32_ref>
b_conv3 RefVariable <tf.Variable 'Variable_5:<...>(128,)
dtype=float32_ref>
b_conv4 RefVariable <tf.Variable 'Variable_7:<...>(256,)
dtype=float32_ref>
b_fc1 RefVariable <tf.Variable 'Variable_9:<...>9216,)
dtype=float32_ref>
b_fc2 RefVariable <tf.Variable 'Variable_11<...>e=(1,)
dtype=float32_ref>
'''