TDA2EVM5777: Error while generating .bin file using Import tool

Dipti Natu

Part Number: TDA2EVM5777

Hi,

While converting Inception_v1 using Import Tool(version 01.01) from PROCESSOR_SDK_VISION_03_04_00_00 there was an error : "Could not find the requested input Data : InceptionV1/InceptionV1/Conv2d_1a_7x7/Relu !! " .

All the instructions given in chapter "3.6.5 Importing Tensorflow Models" in ti_components\algorithms\REL.TIDL.01.01.03.00\modules\ti_dl\docs\TIDeepLearningLibrary_UserGuide.pdf were followed still the issue was not resolved.

The same error was encountered with MobileNet_v1 and PROCESSOR_SDK_VISION_03_07_00_00 .

Please let us know what needs to be done in this case.

Thanks,

Dipti.

over 6 years ago

0 Praveen Eppa1 over 6 years ago

TI__Genius 17580 points

Hi Dipti,

TIDL supports slim based tensorflow models and accepts only optimized frozen graphs.

Did you freeze the graph before giving to import tool ? If not use TF version 1.1 for freezing graph.

Thanks,

Praveen

0 Dipti Natu over 6 years ago in reply to Praveen Eppa1

Intellectual 700 points

Hi Praveen,

Thanks for your help, we realized that when TensorFlow 1.1 is used to create frozen graph issue was resolved.

So, is training the model in TensorFlow 1.1 is mandatory or model can be trained in any version but creating frozen graph in TensorFlow 1.1 is enough?

Thanks,

Dipti

0 Praveen Eppa1 over 6 years ago in reply to Dipti Natu

TI__Genius 17580 points

Hi Dipti,

>> So, is training the model in TensorFlow 1.1 is mandatory or model can be trained in any version but creating frozen graph in TensorFlow 1.1 is enough?

We are not sure about this as we haven't tried this.

Thanks,

Praveen

0 Dipti Natu over 6 years ago in reply to Praveen Eppa1

Intellectual 700 points

Hi Praveen,

We tried importing PSPNet model, which was trained in Tensorflow 1.8 and optimized frozen graph was created in Tensorflow 1.1.

We encountered the following error:

"TF Model File : D:\pb_to_bin\frozen_output_graph_tf1-1.pb

[libprotobuf FATAL C:\Praveen\protobuf-3.2.0rc2\src\google/protobuf/repeated_field.h:1418] CHECK failed: (index) < (current_size_): "

Thanks,

Dipti.

0 Praveen Eppa1 over 6 years ago in reply to Dipti Natu

TI__Genius 17580 points

Hi Dipti,

This could be because of above discussed reason of training in higher TF version and optimizing graph with lower TF version, but not sure so can you share the complete log?

Thanks,

Praveen

0 Dipti Natu over 6 years ago in reply to Praveen Eppa1

Intellectual 700 points

Hi Praveen,

Yes it was because of the version of tensorflow and also because some layers in our model were not supported by TIDL.

Thanks for your help.

Thanks,

Dipti

0 Praveen Eppa1 over 6 years ago in reply to Dipti Natu

TI__Genius 17580 points

Hi Dipti,

Please click on verify answer for the post that you think your query was resolved for closing this thread.

Thanks,

Praveen

Processors

Processors forum

TDA2EVM5777: Error while generating .bin file using Import tool