Revision history [back]

Hi,

I'm trying to export a simple Tensorflow model and import it using opencv.dnn

inp = tf.placeholder(tf.float64, (20,3), name='input')
out = tf.matmul(inp, np.random.random((3,1)),  name='output')

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    tf.saved_model.simple_save(sess, './pb',  {"input": inp},  {"output": out})
    tf.train.write_graph(sess.graph_def, './pbtxt', 'network.pbtxt', True)

and then I try to load it :

net = cv.dnn.readNetFromTensorflow('./pb/saved_model.pb', './pbtxt/network.pbtxt')

I get the following error:

[libprotobuf ERROR opencv/3rdparty/protobuf/src/google/protobuf/wire_format_lite.cc:629] String field 'tensorflow.GradientDef.function_name' contains invalid UTF-8 data when parsing a protocol buffer. Use the 'bytes' type if you intend to send raw bytes. 
Traceback (most recent call last):
     net = cv.dnn.readNetFromTensorflow('./pb/saved_model.pb', './pbtxt/network.pbtxt')
cv2.error: OpenCV(4.0.0-pre) opencv/modules/dnn/src/tensorflow/tf_io.cpp:44: error: (-2:Unspecified error) FAILED: ReadProtoFromBinaryFile(param_file, param). Failed to parse GraphDef file: ./pb/saved_model.pb in function 'ReadTFNetParamsFromBinaryFileOrDie'

Do you have any suggestions what might be the problem and what is the right approach to use the opencv.dnn module?

Thanks!

Hi,

I'm trying to export a simple Tensorflow model and import it using opencv.dnn

inp = tf.placeholder(tf.float64, (20,3), name='input')
out = tf.matmul(inp, np.random.random((3,1)),  name='output')

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    tf.saved_model.simple_save(sess, './pb',  {"input": inp},  {"output": out})
    tf.train.write_graph(sess.graph_def, './pbtxt', 'network.pbtxt', True)

and then I try to load it :

net = cv.dnn.readNetFromTensorflow('./pb/saved_model.pb', './pbtxt/network.pbtxt')

I get the following error:

[libprotobuf ERROR opencv/3rdparty/protobuf/src/google/protobuf/wire_format_lite.cc:629] String field 'tensorflow.GradientDef.function_name' contains invalid UTF-8 data when parsing a protocol buffer. Use the 'bytes' type if you intend to send raw bytes. 
Traceback (most recent call last):
     net = cv.dnn.readNetFromTensorflow('./pb/saved_model.pb', './pbtxt/network.pbtxt')
cv2.error: OpenCV(4.0.0-pre) opencv/modules/dnn/src/tensorflow/tf_io.cpp:44: error: (-2:Unspecified error) FAILED: ReadProtoFromBinaryFile(param_file, param). Failed to parse GraphDef file: ./pb/saved_model.pb in function 'ReadTFNetParamsFromBinaryFileOrDie'

Do you have any suggestions what might be the problem and what is the right approach to use the opencv.dnn module?

Attached is the pbtxt file

Thanks!

  node {
    name: "input"
    op: "Placeholder"
    attr {
      key: "dtype"
      value {
        type: DT_DOUBLE
      }
    }
    attr {
      key: "shape"
      value {
        shape {
          dim {
            size: 20
          }
          dim {
            size: 3
          }
        }
      }
    }
  }
  node {
    name: "output/b"
    op: "Const"
    attr {
      key: "dtype"
      value {
        type: DT_DOUBLE
      }
    }
    attr {
      key: "value"
      value {
        tensor {
          dtype: DT_DOUBLE
          tensor_shape {
            dim {
              size: 3
            }
            dim {
              size: 1
            }
          }
          tensor_content: "@]S\254\275\252\207?\306\234\032\266\244\253\357?\373hw\304\263V\341?"
        }
      }
    }
  }
  node {
    name: "output"
    op: "MatMul"
    input: "input"
    input: "output/b"
    attr {
      key: "T"
      value {
        type: DT_DOUBLE
      }
    }
    attr {
      key: "transpose_a"
      value {
        b: false
      }
    }
    attr {
      key: "transpose_b"
      value {
        b: false
      }
    }
  }
  node {
    name: "init"
    op: "NoOp"
  }
  versions {
    producer: 26
  }