ctc_beam_search_decoder layer is not supported [Tensorflow]

[Medhat__Amr]

Good Morning,

I was trying to convert my own frozen model and it shows me that "ctc_beam_search_decoder" layer is not supported and I have it in my graph, Is there a pre-made custom implementation for this layer that I can use?

Thanks in advance.

[Shubha_R_Intel]

Dear Medhat, Amr,

Here is the Supported Model Optimizer Tensorflow Layers doc. And you are correct, ctc_beam_search_decoder is not on the list for Tensorflow. However, all is not lost. You can make a custom layer to handle it.

Kindly peruse the below documents:

http://docs.openvinotoolkit.org/latest/_docs_MO_DG_prepare_model_customize_model_optimizer_Customize_Model_Optimizer.html

http://docs.openvinotoolkit.org/latest/_docs_IE_DG_Integrate_your_kernels_into_IE.html

Also this IDZ Post which I wrote earlier this year is a bit dated since it has 2018 links but simply go to https://github.com/opencv/dldt and redirect to the 2019 R3 links.

https://software.intel.com/en-us/forums/computer-vision/topic/805980

Hope it helps,

Thanks,

Shubha