• State Resolved
  • Locked Locked
  • Replies 5 replies
  • Answers 2 answers
  • Subscribers 98 subscribers
  • Views 490 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

RTOS: Is there any tensorflow model validated for TIDL OD usecase?

Byeongheon Yoo
Expert 1240 points

Tool/software: TI-RTOS

Hi,

I'm trying to convert Tensorflow model to TIDL model.

The converted TIDL model will be used in TIDL OD usecase. I already tried to convert MobileNet v1 following the TIDL guide documents.

However, i need to test TIDL OD usecase not classification. So, i tried some models which are uploaded in Tensorflow git page but there are lots of unsupported layers type by TIDL.

Could you let me know if you know any validated Tensorflow model for TIDL OD usecase?

How to know a number of layers in pb file when i allocate layersGroupId in tidl_import*.txt file?

Thanks in advance.

Regards

Yoo.

  • 0
    TI__Mastermind 22145 points
    We haven't run any OD models from Tensorflow internally, Some of the users could run.
    You need to run the feature extraction layers in TIDL and rest of the post-processing outside TIDL.
    While running freeze graph script, you need to specify the feature extraction layer as output nodes.
    You can run the import tool with default parameters to set "layersGroupId " initially and then you set the required values in the tidl_import*.txt file and run the import again.
  • 0 in reply to kumar.desappan
    Expert 1240 points
    Ok, i understand.
    By the way, i encountered this error while importing tensorflow ssd-mobile net model.

    Op Type Cast is Not suported will be By passed
    Op Type Shape is Not suported will be By passed
    Op Type Shape is Not suported will be By passed
    Op Type StridedSlice is Not suported will be By passed
    Op Type TensorArrayV3 is Not suported will be By passed
    Op Type Enter is Not suported will be By passed
    Op Type TensorArrayV3 is Not suported will be By passed
    Op Type Enter is Not suported will be By passed

    The reason for this problem is TIDL not support these layer?
    Then how can i convert recent tensorflow model?

    The recent TF version is 1.13 but as i understand TIDL only supports the layer in TF 1.1.0 version.
    It means i can't convert recent TF model if i don't add the definition of unsupported layers in TIDL importing tool.
    AM i right?
  • 0 in reply to Byeongheon Yoo
    TI__Mastermind 22145 points
    Yes, you are right. Currently, we have validated with TF 1.1 only.
    Please use TF 1.1 and also choose box predictors and class prediction convolution layers as output while freezing graph
  • 0 in reply to kumar.desappan
    Expert 1240 points

    Thank you for your suggestion.
    So, you mean add box predictors and class prediction layers when i execute freezing graph?
    Is there any example or guide about adding box predictors and class prediction layer?

    Thank you again.

  • 0 in reply to Byeongheon Yoo
    TI__Mastermind 22145 points
    We don't have example.
    We recommend viewing the SSD model in tensorBoard to find the box predictors and class prediction layer node names .
    Paas the same names as output nodes while running the freezing and optimize graph scripts