Fail convert CTCGreedyDecoder layer to IR with tensorflow

[Tsai__Ward]

I try to convert tensorflow to IR, but CTCGreedyDecoder issue error.

[ 2019-01-29 15:46:01,349 ] [ DEBUG ] [ infer:151 ]  Partial infer for CTCGreedyDecoder
[ 2019-01-29 15:46:01,349 ] [ DEBUG ] [ infer:152 ]  Op: CTCGreedyDecoder
[ 2019-01-29 15:46:01,349 ] [ DEBUG ] [ infer:163 ]  Inputs:
[ 2019-01-29 15:46:01,349 ] [ DEBUG ] [ infer:40 ]  input[0]: shape = [24  1 67], value = <UNKNOWN>
[ 2019-01-29 15:46:01,349 ] [ DEBUG ] [ infer:40 ]  input[1]: shape = [1 1], value = [[24]]
[ 2019-01-29 15:46:01,349 ] [ DEBUG ] [ infer:165 ]  Outputs:
[ 2019-01-29 15:46:01,349 ] [ DEBUG ] [ infer:40 ]  output[0]: shape = [ 1 24  1  1], value = <UNKNOWN>
[ 2019-01-29 15:46:01,349 ] [ DEBUG ] [ infer:40 ]  output[2]: shape = <UNKNOWN>, value = <UNKNOWN>
[ 2019-01-29 15:46:01,349 ] [ DEBUG ] [ infer:40 ]  output[1]: shape = <UNKNOWN>, value = <UNKNOWN>
[ ERROR ]  Shape is not defined for output 2 of "CTCGreedyDecoder".
[ ERROR ]  Shape is not defined for output 1 of "CTCGreedyDecoder".
[ ERROR ]  Cannot infer shapes or values for node "CTCGreedyDecoder".

In lprnet IR, there only one output in CTCGreedyDecoder in LPRNet.xml:

<layer id="78" name="decode" precision="FP16" type="CTCGreedyDecoder">
    <data ctc_merge_repeated="1"/>
    <input>
        <port id="0">
            <dim>88</dim>
            <dim>1</dim>
            <dim>71</dim>
        </port>
        <port id="1">
            <dim>88</dim>
            <dim>1</dim>
        </port>
    </input>
    <output>
        <port id="2">
            <dim>1</dim>
            <dim>88</dim>
            <dim>1</dim>
            <dim>1</dim>
        </port>
    </output>
</layer>

The tf model:

https://drive.google.com/open?id=1sQ9mzf3acwSO1grDC7q81qX6aCPXshIp

Please help. Thank you!