lstm convert error

[thomaszheng]

Describe the bug

A clear and concise description of what the bug is. ocr rec model crnn+lstm and convert to onnx.

ValueError: in user code:

    /Users/xiaomi/ml_python/onnx-tensorflow/onnx_tf/backend_tf_module.py:99 __call__  *
        output_ops = self.backend._onnx_node_to_tensorflow_op(onnx_node,
    /Users/xiaomi/ml_python/onnx-tensorflow/onnx_tf/backend.py:347 _onnx_node_to_tensorflow_op  *
        return handler.handle(node, tensor_dict=tensor_dict, strict=strict)
    /Users/xiaomi/ml_python/onnx-tensorflow/onnx_tf/handlers/handler.py:59 handle  *
        return ver_handle(node, **kwargs)
    /Users/xiaomi/ml_python/onnx-tensorflow/onnx_tf/handlers/backend/lstm.py:287 version_7  *
        return cls._common(node, **kwargs)
    /Users/xiaomi/ml_python/onnx-tensorflow/onnx_tf/handlers/backend/lstm.py:249 _common  *
        outputs, states = cls.rnn(x, tf.compat.v1.nn.rnn_cell.LSTMCell,
    /Users/xiaomi/ml_python/onnx-tensorflow/onnx_tf/handlers/backend/rnn_mixin.py:49 rnn  *
        outputs, states = tf.compat.v1.nn.bidirectional_dynamic_rnn(
    /usr/local/lib/python3.8/site-packages/tensorflow/python/util/deprecation.py:346 new_func  **
        return func(*args, **kwargs)
    /usr/local/lib/python3.8/site-packages/tensorflow/python/util/dispatch.py:206 wrapper
        return target(*args, **kwargs)
    /usr/local/lib/python3.8/site-packages/tensorflow/python/ops/rnn.py:438 bidirectional_dynamic_rnn
        output_fw, output_state_fw = dynamic_rnn(
    /usr/local/lib/python3.8/site-packages/tensorflow/python/util/deprecation.py:346 new_func
        return func(*args, **kwargs)
    /usr/local/lib/python3.8/site-packages/tensorflow/python/util/dispatch.py:206 wrapper
        return target(*args, **kwargs)
    /usr/local/lib/python3.8/site-packages/tensorflow/python/ops/rnn.py:684 dynamic_rnn
        (outputs, final_state) = _dynamic_rnn_loop(
    /usr/local/lib/python3.8/site-packages/tensorflow/python/ops/rnn.py:764 _dynamic_rnn_loop
        raise ValueError(

    ValueError: Input size (depth of inputs) must be accessible via shape inference, but saw value None.

To Reproduce

Please give us instructions to reproduce your problem.

onnx-tf convert -i rec.onnx -o rec_n --logging_level DEBU

ONNX model file

rec.onnx

Python, ONNX, ONNX-TF, Tensorflow version

This section can be obtained by running get_version.py from util folder.

• Python version: 3.8.6
• ONNX version: 1.9.0
• ONNX-TF version:1.9.0
• Tensorflow version: 2.6.0

Additional context

if i change file rnn_mixin.py as fallow, it can convert without error, but the model can't inference

    if direction == "forward":
      outputs, states = tf.compat.v1.nn.dynamic_rnn(cell_fw, x, **rnn_kwargs)
    elif direction == "bidirectional":
      print(x.shape)
      x.set_shape([None, None, 288])
      outputs, states = tf.compat.v1.nn.bidirectional_dynamic_rnn(
          cell_fw, cell_bw, x, **rnn_kwargs)