Inferencing Semantic segmentation models in tensorflow on TDA 2X / 3X / 4X

Prajakta Kulkarni

Hi,

Is there a way to port a semantic segmentation model trained in tensorflow on TDA 2X / 3X / 4X?

While trying the experiment out, I have already observed following problems:

1. Tensorflow version supported by TIDL is pretty old i.e. 1.1. Many networks currently are using tensorflow 1.8 or even higher upto 1.13.

2. Semantic segmentation requires transposed convolution layer which is not supported by TIDL in tensorflow.

Is there any enhancement on tensorflow higher version support?

If transposed convolution is not supported what are the ways to deploy a segmentation model in tensorflow on TDA2X / 3X / 4X?

Thanks in advance,

Prajakta

0 Praveen Eppa1 over 6 years ago

Hi Prajakta,

TIDL 00.09 release on TDA4X is validated with tensorflow version 1.12.

For TDA2X/3X the support for tensorflow version 1.12 is in progress and this support will be part of next TIDL 01.04 release.

You can try semantic segmentation on TDA4X as DeConvolution is supported in the tenosrflow import.

Thanks,

Praveen

0 Prajakta Kulkarni over 6 years ago in reply to Praveen Eppa1

Hi Praveen,

Thank you so much for this clarification.

Can you also let me know how can I get access to SDK for TDA 4X?

Prajakta

0 Praveen Eppa1 over 6 years ago in reply to Prajakta Kulkarni

Hi Prajakta,

I have forwarded your query internally and waiting for their response.

Thanks for your patience.

Regards,

Praveen

0 Praveen Eppa1 over 5 years ago in reply to Praveen Eppa1

Hi Prajakta,

I would recommend you to contact local TI FAE, they will suggest you on the next steps.

Thanks,

Praveen

Processors forum