Compiler: TDA2X - Tensorflow Model Layer-By-Layer matching

Akshat Agrawal1

Tool/software: TI C/C++ Compiler

Hi,

We are trying to import a Tensorflow customized Object Detection model on TIDL Import tool (REL.TIDL.01.01.03.00). We are able to achieve .net and .params files. We aren't able to achieve the expected results on the output layer, so as suggested by TI we are trying to do layer-by-layer matching with the Tensorflow model.

In our Tensorflow model, we normalize the model input between "0.0" to "1.0" by dividing the image by "255.0". The layer we are trying to match is a TIDL_ConvolutionLayer which is Depth wise separable convolution with Batch Normalization and ReLU. We have used the following parameters in our config file -

inElementType = 0

inQuantFactor = 255 and inQuantFactor = -1

Also we scale the values with the obtained QValues (11 in case of inQuantFactor = 255 and 839 in case of inQuantFactor = -1) an then compare the output with the Tensorflow's float32 output. We are not able to match the layer outputs.

Please review our process and let us know if we are doing something incorrect.

Thanks and Regards,

Akshat Agrawal.

over 5 years ago

0 Praveen Eppa1 over 5 years ago

TI__Genius 17580 points

Hi Akshat,

Could you please try with import tool from REL.TIDL.01.01.04.00 and check ?

Thanks,

Praveen

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

TI__Genius 17580 points

Hi Akshat,

I haven't heard back from you, I'm assuming you were able to resolve your issue.

If not, just post a reply below (or create a new thread if the thread has locked due to time-out).

Thanks,

Praveen

0 Vyom Mishra1 over 5 years ago in reply to Praveen Eppa1

Genius 4590 points

Hi Anshu,

We have tried out the latest Release of TIDL_01_01_04_00 and modified the tool as per our Model requirements.

The model got imported successfully.

**Our model has No Post-processing included in it.

Currently, we are matching the layer-by-layer output of the traces with PC output.

Can you please suggest us to get the traces in floating point as we get in TDA4x by adding the configuration parameters "writeTraceLevel = 3 "

Thanks and Regards,

Vyom Mishra

0 Praveen Eppa1 over 5 years ago in reply to Vyom Mishra1

TI__Genius 17580 points

Hi Vyom,

You can dump floating point output from TIDL by defining the below macro in tidl_tb.c file.

#define ENABLE_FLOAT_TRACE (1)

Thanks,

Praveen

Processors

Processors forum

Compiler: TDA2X - Tensorflow Model Layer-By-Layer matching